1 //===-- APFloat.cpp - Implement APFloat class -----------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements a class to represent arbitrary precision floating 11 // point values and provide a variety of arithmetic operations on them. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "llvm/ADT/APFloat.h" 16 #include "llvm/ADT/APSInt.h" 17 #include "llvm/ADT/FoldingSet.h" 18 #include "llvm/ADT/Hashing.h" 19 #include "llvm/ADT/StringExtras.h" 20 #include "llvm/ADT/StringRef.h" 21 #include "llvm/Support/ErrorHandling.h" 22 #include "llvm/Support/MathExtras.h" 23 #include <cstring> 24 #include <limits.h> 25 26 using namespace llvm; 27 28 /// A macro used to combine two fcCategory enums into one key which can be used 29 /// in a switch statement to classify how the interaction of two APFloat's 30 /// categories affects an operation. 31 /// 32 /// TODO: If clang source code is ever allowed to use constexpr in its own 33 /// codebase, change this into a static inline function. 34 #define PackCategoriesIntoKey(_lhs, _rhs) ((_lhs) * 4 + (_rhs)) 35 36 /* Assumed in hexadecimal significand parsing, and conversion to 37 hexadecimal strings. */ 38 #define COMPILE_TIME_ASSERT(cond) extern int CTAssert[(cond) ? 1 : -1] 39 COMPILE_TIME_ASSERT(integerPartWidth % 4 == 0); 40 41 namespace llvm { 42 43 /* Represents floating point arithmetic semantics. */ 44 struct fltSemantics { 45 /* The largest E such that 2^E is representable; this matches the 46 definition of IEEE 754. */ 47 APFloat::ExponentType maxExponent; 48 49 /* The smallest E such that 2^E is a normalized number; this 50 matches the definition of IEEE 754. */ 51 APFloat::ExponentType minExponent; 52 53 /* Number of bits in the significand. This includes the integer 54 bit. */ 55 unsigned int precision; 56 }; 57 58 const fltSemantics APFloat::IEEEhalf = { 15, -14, 11 }; 59 const fltSemantics APFloat::IEEEsingle = { 127, -126, 24 }; 60 const fltSemantics APFloat::IEEEdouble = { 1023, -1022, 53 }; 61 const fltSemantics APFloat::IEEEquad = { 16383, -16382, 113 }; 62 const fltSemantics APFloat::x87DoubleExtended = { 16383, -16382, 64 }; 63 const fltSemantics APFloat::Bogus = { 0, 0, 0 }; 64 65 /* The PowerPC format consists of two doubles. It does not map cleanly 66 onto the usual format above. It is approximated using twice the 67 mantissa bits. Note that for exponents near the double minimum, 68 we no longer can represent the full 106 mantissa bits, so those 69 will be treated as denormal numbers. 70 71 FIXME: While this approximation is equivalent to what GCC uses for 72 compile-time arithmetic on PPC double-double numbers, it is not able 73 to represent all possible values held by a PPC double-double number, 74 for example: (long double) 1.0 + (long double) 0x1p-106 75 Should this be replaced by a full emulation of PPC double-double? */ 76 const fltSemantics APFloat::PPCDoubleDouble = { 1023, -1022 + 53, 53 + 53 }; 77 78 /* A tight upper bound on number of parts required to hold the value 79 pow(5, power) is 80 81 power * 815 / (351 * integerPartWidth) + 1 82 83 However, whilst the result may require only this many parts, 84 because we are multiplying two values to get it, the 85 multiplication may require an extra part with the excess part 86 being zero (consider the trivial case of 1 * 1, tcFullMultiply 87 requires two parts to hold the single-part result). So we add an 88 extra one to guarantee enough space whilst multiplying. */ 89 const unsigned int maxExponent = 16383; 90 const unsigned int maxPrecision = 113; 91 const unsigned int maxPowerOfFiveExponent = maxExponent + maxPrecision - 1; 92 const unsigned int maxPowerOfFiveParts = 2 + ((maxPowerOfFiveExponent * 815) 93 / (351 * integerPartWidth)); 94 } 95 96 /* A bunch of private, handy routines. */ 97 98 static inline unsigned int 99 partCountForBits(unsigned int bits) 100 { 101 return ((bits) + integerPartWidth - 1) / integerPartWidth; 102 } 103 104 /* Returns 0U-9U. Return values >= 10U are not digits. */ 105 static inline unsigned int 106 decDigitValue(unsigned int c) 107 { 108 return c - '0'; 109 } 110 111 /* Return the value of a decimal exponent of the form 112 [+-]ddddddd. 113 114 If the exponent overflows, returns a large exponent with the 115 appropriate sign. */ 116 static int 117 readExponent(StringRef::iterator begin, StringRef::iterator end) 118 { 119 bool isNegative; 120 unsigned int absExponent; 121 const unsigned int overlargeExponent = 24000; /* FIXME. */ 122 StringRef::iterator p = begin; 123 124 assert(p != end && "Exponent has no digits"); 125 126 isNegative = (*p == '-'); 127 if (*p == '-' || *p == '+') { 128 p++; 129 assert(p != end && "Exponent has no digits"); 130 } 131 132 absExponent = decDigitValue(*p++); 133 assert(absExponent < 10U && "Invalid character in exponent"); 134 135 for (; p != end; ++p) { 136 unsigned int value; 137 138 value = decDigitValue(*p); 139 assert(value < 10U && "Invalid character in exponent"); 140 141 value += absExponent * 10; 142 if (absExponent >= overlargeExponent) { 143 absExponent = overlargeExponent; 144 p = end; /* outwit assert below */ 145 break; 146 } 147 absExponent = value; 148 } 149 150 assert(p == end && "Invalid exponent in exponent"); 151 152 if (isNegative) 153 return -(int) absExponent; 154 else 155 return (int) absExponent; 156 } 157 158 /* This is ugly and needs cleaning up, but I don't immediately see 159 how whilst remaining safe. */ 160 static int 161 totalExponent(StringRef::iterator p, StringRef::iterator end, 162 int exponentAdjustment) 163 { 164 int unsignedExponent; 165 bool negative, overflow; 166 int exponent = 0; 167 168 assert(p != end && "Exponent has no digits"); 169 170 negative = *p == '-'; 171 if (*p == '-' || *p == '+') { 172 p++; 173 assert(p != end && "Exponent has no digits"); 174 } 175 176 unsignedExponent = 0; 177 overflow = false; 178 for (; p != end; ++p) { 179 unsigned int value; 180 181 value = decDigitValue(*p); 182 assert(value < 10U && "Invalid character in exponent"); 183 184 unsignedExponent = unsignedExponent * 10 + value; 185 if (unsignedExponent > 32767) { 186 overflow = true; 187 break; 188 } 189 } 190 191 if (exponentAdjustment > 32767 || exponentAdjustment < -32768) 192 overflow = true; 193 194 if (!overflow) { 195 exponent = unsignedExponent; 196 if (negative) 197 exponent = -exponent; 198 exponent += exponentAdjustment; 199 if (exponent > 32767 || exponent < -32768) 200 overflow = true; 201 } 202 203 if (overflow) 204 exponent = negative ? -32768: 32767; 205 206 return exponent; 207 } 208 209 static StringRef::iterator 210 skipLeadingZeroesAndAnyDot(StringRef::iterator begin, StringRef::iterator end, 211 StringRef::iterator *dot) 212 { 213 StringRef::iterator p = begin; 214 *dot = end; 215 while (*p == '0' && p != end) 216 p++; 217 218 if (*p == '.') { 219 *dot = p++; 220 221 assert(end - begin != 1 && "Significand has no digits"); 222 223 while (*p == '0' && p != end) 224 p++; 225 } 226 227 return p; 228 } 229 230 /* Given a normal decimal floating point number of the form 231 232 dddd.dddd[eE][+-]ddd 233 234 where the decimal point and exponent are optional, fill out the 235 structure D. Exponent is appropriate if the significand is 236 treated as an integer, and normalizedExponent if the significand 237 is taken to have the decimal point after a single leading 238 non-zero digit. 239 240 If the value is zero, V->firstSigDigit points to a non-digit, and 241 the return exponent is zero. 242 */ 243 struct decimalInfo { 244 const char *firstSigDigit; 245 const char *lastSigDigit; 246 int exponent; 247 int normalizedExponent; 248 }; 249 250 static void 251 interpretDecimal(StringRef::iterator begin, StringRef::iterator end, 252 decimalInfo *D) 253 { 254 StringRef::iterator dot = end; 255 StringRef::iterator p = skipLeadingZeroesAndAnyDot (begin, end, &dot); 256 257 D->firstSigDigit = p; 258 D->exponent = 0; 259 D->normalizedExponent = 0; 260 261 for (; p != end; ++p) { 262 if (*p == '.') { 263 assert(dot == end && "String contains multiple dots"); 264 dot = p++; 265 if (p == end) 266 break; 267 } 268 if (decDigitValue(*p) >= 10U) 269 break; 270 } 271 272 if (p != end) { 273 assert((*p == 'e' || *p == 'E') && "Invalid character in significand"); 274 assert(p != begin && "Significand has no digits"); 275 assert((dot == end || p - begin != 1) && "Significand has no digits"); 276 277 /* p points to the first non-digit in the string */ 278 D->exponent = readExponent(p + 1, end); 279 280 /* Implied decimal point? */ 281 if (dot == end) 282 dot = p; 283 } 284 285 /* If number is all zeroes accept any exponent. */ 286 if (p != D->firstSigDigit) { 287 /* Drop insignificant trailing zeroes. */ 288 if (p != begin) { 289 do 290 do 291 p--; 292 while (p != begin && *p == '0'); 293 while (p != begin && *p == '.'); 294 } 295 296 /* Adjust the exponents for any decimal point. */ 297 D->exponent += static_cast<APFloat::ExponentType>((dot - p) - (dot > p)); 298 D->normalizedExponent = (D->exponent + 299 static_cast<APFloat::ExponentType>((p - D->firstSigDigit) 300 - (dot > D->firstSigDigit && dot < p))); 301 } 302 303 D->lastSigDigit = p; 304 } 305 306 /* Return the trailing fraction of a hexadecimal number. 307 DIGITVALUE is the first hex digit of the fraction, P points to 308 the next digit. */ 309 static lostFraction 310 trailingHexadecimalFraction(StringRef::iterator p, StringRef::iterator end, 311 unsigned int digitValue) 312 { 313 unsigned int hexDigit; 314 315 /* If the first trailing digit isn't 0 or 8 we can work out the 316 fraction immediately. */ 317 if (digitValue > 8) 318 return lfMoreThanHalf; 319 else if (digitValue < 8 && digitValue > 0) 320 return lfLessThanHalf; 321 322 // Otherwise we need to find the first non-zero digit. 323 while (p != end && (*p == '0' || *p == '.')) 324 p++; 325 326 assert(p != end && "Invalid trailing hexadecimal fraction!"); 327 328 hexDigit = hexDigitValue(*p); 329 330 /* If we ran off the end it is exactly zero or one-half, otherwise 331 a little more. */ 332 if (hexDigit == -1U) 333 return digitValue == 0 ? lfExactlyZero: lfExactlyHalf; 334 else 335 return digitValue == 0 ? lfLessThanHalf: lfMoreThanHalf; 336 } 337 338 /* Return the fraction lost were a bignum truncated losing the least 339 significant BITS bits. */ 340 static lostFraction 341 lostFractionThroughTruncation(const integerPart *parts, 342 unsigned int partCount, 343 unsigned int bits) 344 { 345 unsigned int lsb; 346 347 lsb = APInt::tcLSB(parts, partCount); 348 349 /* Note this is guaranteed true if bits == 0, or LSB == -1U. */ 350 if (bits <= lsb) 351 return lfExactlyZero; 352 if (bits == lsb + 1) 353 return lfExactlyHalf; 354 if (bits <= partCount * integerPartWidth && 355 APInt::tcExtractBit(parts, bits - 1)) 356 return lfMoreThanHalf; 357 358 return lfLessThanHalf; 359 } 360 361 /* Shift DST right BITS bits noting lost fraction. */ 362 static lostFraction 363 shiftRight(integerPart *dst, unsigned int parts, unsigned int bits) 364 { 365 lostFraction lost_fraction; 366 367 lost_fraction = lostFractionThroughTruncation(dst, parts, bits); 368 369 APInt::tcShiftRight(dst, parts, bits); 370 371 return lost_fraction; 372 } 373 374 /* Combine the effect of two lost fractions. */ 375 static lostFraction 376 combineLostFractions(lostFraction moreSignificant, 377 lostFraction lessSignificant) 378 { 379 if (lessSignificant != lfExactlyZero) { 380 if (moreSignificant == lfExactlyZero) 381 moreSignificant = lfLessThanHalf; 382 else if (moreSignificant == lfExactlyHalf) 383 moreSignificant = lfMoreThanHalf; 384 } 385 386 return moreSignificant; 387 } 388 389 /* The error from the true value, in half-ulps, on multiplying two 390 floating point numbers, which differ from the value they 391 approximate by at most HUE1 and HUE2 half-ulps, is strictly less 392 than the returned value. 393 394 See "How to Read Floating Point Numbers Accurately" by William D 395 Clinger. */ 396 static unsigned int 397 HUerrBound(bool inexactMultiply, unsigned int HUerr1, unsigned int HUerr2) 398 { 399 assert(HUerr1 < 2 || HUerr2 < 2 || (HUerr1 + HUerr2 < 8)); 400 401 if (HUerr1 + HUerr2 == 0) 402 return inexactMultiply * 2; /* <= inexactMultiply half-ulps. */ 403 else 404 return inexactMultiply + 2 * (HUerr1 + HUerr2); 405 } 406 407 /* The number of ulps from the boundary (zero, or half if ISNEAREST) 408 when the least significant BITS are truncated. BITS cannot be 409 zero. */ 410 static integerPart 411 ulpsFromBoundary(const integerPart *parts, unsigned int bits, bool isNearest) 412 { 413 unsigned int count, partBits; 414 integerPart part, boundary; 415 416 assert(bits != 0); 417 418 bits--; 419 count = bits / integerPartWidth; 420 partBits = bits % integerPartWidth + 1; 421 422 part = parts[count] & (~(integerPart) 0 >> (integerPartWidth - partBits)); 423 424 if (isNearest) 425 boundary = (integerPart) 1 << (partBits - 1); 426 else 427 boundary = 0; 428 429 if (count == 0) { 430 if (part - boundary <= boundary - part) 431 return part - boundary; 432 else 433 return boundary - part; 434 } 435 436 if (part == boundary) { 437 while (--count) 438 if (parts[count]) 439 return ~(integerPart) 0; /* A lot. */ 440 441 return parts[0]; 442 } else if (part == boundary - 1) { 443 while (--count) 444 if (~parts[count]) 445 return ~(integerPart) 0; /* A lot. */ 446 447 return -parts[0]; 448 } 449 450 return ~(integerPart) 0; /* A lot. */ 451 } 452 453 /* Place pow(5, power) in DST, and return the number of parts used. 454 DST must be at least one part larger than size of the answer. */ 455 static unsigned int 456 powerOf5(integerPart *dst, unsigned int power) 457 { 458 static const integerPart firstEightPowers[] = { 1, 5, 25, 125, 625, 3125, 459 15625, 78125 }; 460 integerPart pow5s[maxPowerOfFiveParts * 2 + 5]; 461 pow5s[0] = 78125 * 5; 462 463 unsigned int partsCount[16] = { 1 }; 464 integerPart scratch[maxPowerOfFiveParts], *p1, *p2, *pow5; 465 unsigned int result; 466 assert(power <= maxExponent); 467 468 p1 = dst; 469 p2 = scratch; 470 471 *p1 = firstEightPowers[power & 7]; 472 power >>= 3; 473 474 result = 1; 475 pow5 = pow5s; 476 477 for (unsigned int n = 0; power; power >>= 1, n++) { 478 unsigned int pc; 479 480 pc = partsCount[n]; 481 482 /* Calculate pow(5,pow(2,n+3)) if we haven't yet. */ 483 if (pc == 0) { 484 pc = partsCount[n - 1]; 485 APInt::tcFullMultiply(pow5, pow5 - pc, pow5 - pc, pc, pc); 486 pc *= 2; 487 if (pow5[pc - 1] == 0) 488 pc--; 489 partsCount[n] = pc; 490 } 491 492 if (power & 1) { 493 integerPart *tmp; 494 495 APInt::tcFullMultiply(p2, p1, pow5, result, pc); 496 result += pc; 497 if (p2[result - 1] == 0) 498 result--; 499 500 /* Now result is in p1 with partsCount parts and p2 is scratch 501 space. */ 502 tmp = p1, p1 = p2, p2 = tmp; 503 } 504 505 pow5 += pc; 506 } 507 508 if (p1 != dst) 509 APInt::tcAssign(dst, p1, result); 510 511 return result; 512 } 513 514 /* Zero at the end to avoid modular arithmetic when adding one; used 515 when rounding up during hexadecimal output. */ 516 static const char hexDigitsLower[] = "0123456789abcdef0"; 517 static const char hexDigitsUpper[] = "0123456789ABCDEF0"; 518 static const char infinityL[] = "infinity"; 519 static const char infinityU[] = "INFINITY"; 520 static const char NaNL[] = "nan"; 521 static const char NaNU[] = "NAN"; 522 523 /* Write out an integerPart in hexadecimal, starting with the most 524 significant nibble. Write out exactly COUNT hexdigits, return 525 COUNT. */ 526 static unsigned int 527 partAsHex (char *dst, integerPart part, unsigned int count, 528 const char *hexDigitChars) 529 { 530 unsigned int result = count; 531 532 assert(count != 0 && count <= integerPartWidth / 4); 533 534 part >>= (integerPartWidth - 4 * count); 535 while (count--) { 536 dst[count] = hexDigitChars[part & 0xf]; 537 part >>= 4; 538 } 539 540 return result; 541 } 542 543 /* Write out an unsigned decimal integer. */ 544 static char * 545 writeUnsignedDecimal (char *dst, unsigned int n) 546 { 547 char buff[40], *p; 548 549 p = buff; 550 do 551 *p++ = '0' + n % 10; 552 while (n /= 10); 553 554 do 555 *dst++ = *--p; 556 while (p != buff); 557 558 return dst; 559 } 560 561 /* Write out a signed decimal integer. */ 562 static char * 563 writeSignedDecimal (char *dst, int value) 564 { 565 if (value < 0) { 566 *dst++ = '-'; 567 dst = writeUnsignedDecimal(dst, -(unsigned) value); 568 } else 569 dst = writeUnsignedDecimal(dst, value); 570 571 return dst; 572 } 573 574 /* Constructors. */ 575 void 576 APFloat::initialize(const fltSemantics *ourSemantics) 577 { 578 unsigned int count; 579 580 semantics = ourSemantics; 581 count = partCount(); 582 if (count > 1) 583 significand.parts = new integerPart[count]; 584 } 585 586 void 587 APFloat::freeSignificand() 588 { 589 if (needsCleanup()) 590 delete [] significand.parts; 591 } 592 593 void 594 APFloat::assign(const APFloat &rhs) 595 { 596 assert(semantics == rhs.semantics); 597 598 sign = rhs.sign; 599 category = rhs.category; 600 exponent = rhs.exponent; 601 if (isFiniteNonZero() || category == fcNaN) 602 copySignificand(rhs); 603 } 604 605 void 606 APFloat::copySignificand(const APFloat &rhs) 607 { 608 assert(isFiniteNonZero() || category == fcNaN); 609 assert(rhs.partCount() >= partCount()); 610 611 APInt::tcAssign(significandParts(), rhs.significandParts(), 612 partCount()); 613 } 614 615 /* Make this number a NaN, with an arbitrary but deterministic value 616 for the significand. If double or longer, this is a signalling NaN, 617 which may not be ideal. If float, this is QNaN(0). */ 618 void APFloat::makeNaN(bool SNaN, bool Negative, const APInt *fill) 619 { 620 category = fcNaN; 621 sign = Negative; 622 623 integerPart *significand = significandParts(); 624 unsigned numParts = partCount(); 625 626 // Set the significand bits to the fill. 627 if (!fill || fill->getNumWords() < numParts) 628 APInt::tcSet(significand, 0, numParts); 629 if (fill) { 630 APInt::tcAssign(significand, fill->getRawData(), 631 std::min(fill->getNumWords(), numParts)); 632 633 // Zero out the excess bits of the significand. 634 unsigned bitsToPreserve = semantics->precision - 1; 635 unsigned part = bitsToPreserve / 64; 636 bitsToPreserve %= 64; 637 significand[part] &= ((1ULL << bitsToPreserve) - 1); 638 for (part++; part != numParts; ++part) 639 significand[part] = 0; 640 } 641 642 unsigned QNaNBit = semantics->precision - 2; 643 644 if (SNaN) { 645 // We always have to clear the QNaN bit to make it an SNaN. 646 APInt::tcClearBit(significand, QNaNBit); 647 648 // If there are no bits set in the payload, we have to set 649 // *something* to make it a NaN instead of an infinity; 650 // conventionally, this is the next bit down from the QNaN bit. 651 if (APInt::tcIsZero(significand, numParts)) 652 APInt::tcSetBit(significand, QNaNBit - 1); 653 } else { 654 // We always have to set the QNaN bit to make it a QNaN. 655 APInt::tcSetBit(significand, QNaNBit); 656 } 657 658 // For x87 extended precision, we want to make a NaN, not a 659 // pseudo-NaN. Maybe we should expose the ability to make 660 // pseudo-NaNs? 661 if (semantics == &APFloat::x87DoubleExtended) 662 APInt::tcSetBit(significand, QNaNBit + 1); 663 } 664 665 APFloat APFloat::makeNaN(const fltSemantics &Sem, bool SNaN, bool Negative, 666 const APInt *fill) { 667 APFloat value(Sem, uninitialized); 668 value.makeNaN(SNaN, Negative, fill); 669 return value; 670 } 671 672 APFloat & 673 APFloat::operator=(const APFloat &rhs) 674 { 675 if (this != &rhs) { 676 if (semantics != rhs.semantics) { 677 freeSignificand(); 678 initialize(rhs.semantics); 679 } 680 assign(rhs); 681 } 682 683 return *this; 684 } 685 686 bool 687 APFloat::isDenormal() const { 688 return isFiniteNonZero() && (exponent == semantics->minExponent) && 689 (APInt::tcExtractBit(significandParts(), 690 semantics->precision - 1) == 0); 691 } 692 693 bool 694 APFloat::isSmallest() const { 695 // The smallest number by magnitude in our format will be the smallest 696 // denormal, i.e. the floating point number with exponent being minimum 697 // exponent and significand bitwise equal to 1 (i.e. with MSB equal to 0). 698 return isFiniteNonZero() && exponent == semantics->minExponent && 699 significandMSB() == 0; 700 } 701 702 bool APFloat::isSignificandAllOnes() const { 703 // Test if the significand excluding the integral bit is all ones. This allows 704 // us to test for binade boundaries. 705 const integerPart *Parts = significandParts(); 706 const unsigned PartCount = partCount(); 707 for (unsigned i = 0; i < PartCount - 1; i++) 708 if (~Parts[i]) 709 return false; 710 711 // Set the unused high bits to all ones when we compare. 712 const unsigned NumHighBits = 713 PartCount*integerPartWidth - semantics->precision + 1; 714 assert(NumHighBits <= integerPartWidth && "Can not have more high bits to " 715 "fill than integerPartWidth"); 716 const integerPart HighBitFill = 717 ~integerPart(0) << (integerPartWidth - NumHighBits); 718 if (~(Parts[PartCount - 1] | HighBitFill)) 719 return false; 720 721 return true; 722 } 723 724 bool APFloat::isSignificandAllZeros() const { 725 // Test if the significand excluding the integral bit is all zeros. This 726 // allows us to test for binade boundaries. 727 const integerPart *Parts = significandParts(); 728 const unsigned PartCount = partCount(); 729 730 for (unsigned i = 0; i < PartCount - 1; i++) 731 if (Parts[i]) 732 return false; 733 734 const unsigned NumHighBits = 735 PartCount*integerPartWidth - semantics->precision + 1; 736 assert(NumHighBits <= integerPartWidth && "Can not have more high bits to " 737 "clear than integerPartWidth"); 738 const integerPart HighBitMask = ~integerPart(0) >> NumHighBits; 739 740 if (Parts[PartCount - 1] & HighBitMask) 741 return false; 742 743 return true; 744 } 745 746 bool 747 APFloat::isLargest() const { 748 // The largest number by magnitude in our format will be the floating point 749 // number with maximum exponent and with significand that is all ones. 750 return isFiniteNonZero() && exponent == semantics->maxExponent 751 && isSignificandAllOnes(); 752 } 753 754 bool 755 APFloat::bitwiseIsEqual(const APFloat &rhs) const { 756 if (this == &rhs) 757 return true; 758 if (semantics != rhs.semantics || 759 category != rhs.category || 760 sign != rhs.sign) 761 return false; 762 if (category==fcZero || category==fcInfinity) 763 return true; 764 else if (isFiniteNonZero() && exponent!=rhs.exponent) 765 return false; 766 else { 767 int i= partCount(); 768 const integerPart* p=significandParts(); 769 const integerPart* q=rhs.significandParts(); 770 for (; i>0; i--, p++, q++) { 771 if (*p != *q) 772 return false; 773 } 774 return true; 775 } 776 } 777 778 APFloat::APFloat(const fltSemantics &ourSemantics, integerPart value) { 779 initialize(&ourSemantics); 780 sign = 0; 781 zeroSignificand(); 782 exponent = ourSemantics.precision - 1; 783 significandParts()[0] = value; 784 normalize(rmNearestTiesToEven, lfExactlyZero); 785 } 786 787 APFloat::APFloat(const fltSemantics &ourSemantics) { 788 initialize(&ourSemantics); 789 category = fcZero; 790 sign = false; 791 } 792 793 APFloat::APFloat(const fltSemantics &ourSemantics, uninitializedTag tag) { 794 // Allocates storage if necessary but does not initialize it. 795 initialize(&ourSemantics); 796 } 797 798 APFloat::APFloat(const fltSemantics &ourSemantics, StringRef text) { 799 initialize(&ourSemantics); 800 convertFromString(text, rmNearestTiesToEven); 801 } 802 803 APFloat::APFloat(const APFloat &rhs) { 804 initialize(rhs.semantics); 805 assign(rhs); 806 } 807 808 APFloat::~APFloat() 809 { 810 freeSignificand(); 811 } 812 813 // Profile - This method 'profiles' an APFloat for use with FoldingSet. 814 void APFloat::Profile(FoldingSetNodeID& ID) const { 815 ID.Add(bitcastToAPInt()); 816 } 817 818 unsigned int 819 APFloat::partCount() const 820 { 821 return partCountForBits(semantics->precision + 1); 822 } 823 824 unsigned int 825 APFloat::semanticsPrecision(const fltSemantics &semantics) 826 { 827 return semantics.precision; 828 } 829 830 const integerPart * 831 APFloat::significandParts() const 832 { 833 return const_cast<APFloat *>(this)->significandParts(); 834 } 835 836 integerPart * 837 APFloat::significandParts() 838 { 839 if (partCount() > 1) 840 return significand.parts; 841 else 842 return &significand.part; 843 } 844 845 void 846 APFloat::zeroSignificand() 847 { 848 category = fcNormal; 849 APInt::tcSet(significandParts(), 0, partCount()); 850 } 851 852 /* Increment an fcNormal floating point number's significand. */ 853 void 854 APFloat::incrementSignificand() 855 { 856 integerPart carry; 857 858 carry = APInt::tcIncrement(significandParts(), partCount()); 859 860 /* Our callers should never cause us to overflow. */ 861 assert(carry == 0); 862 (void)carry; 863 } 864 865 /* Add the significand of the RHS. Returns the carry flag. */ 866 integerPart 867 APFloat::addSignificand(const APFloat &rhs) 868 { 869 integerPart *parts; 870 871 parts = significandParts(); 872 873 assert(semantics == rhs.semantics); 874 assert(exponent == rhs.exponent); 875 876 return APInt::tcAdd(parts, rhs.significandParts(), 0, partCount()); 877 } 878 879 /* Subtract the significand of the RHS with a borrow flag. Returns 880 the borrow flag. */ 881 integerPart 882 APFloat::subtractSignificand(const APFloat &rhs, integerPart borrow) 883 { 884 integerPart *parts; 885 886 parts = significandParts(); 887 888 assert(semantics == rhs.semantics); 889 assert(exponent == rhs.exponent); 890 891 return APInt::tcSubtract(parts, rhs.significandParts(), borrow, 892 partCount()); 893 } 894 895 /* Multiply the significand of the RHS. If ADDEND is non-NULL, add it 896 on to the full-precision result of the multiplication. Returns the 897 lost fraction. */ 898 lostFraction 899 APFloat::multiplySignificand(const APFloat &rhs, const APFloat *addend) 900 { 901 unsigned int omsb; // One, not zero, based MSB. 902 unsigned int partsCount, newPartsCount, precision; 903 integerPart *lhsSignificand; 904 integerPart scratch[4]; 905 integerPart *fullSignificand; 906 lostFraction lost_fraction; 907 bool ignored; 908 909 assert(semantics == rhs.semantics); 910 911 precision = semantics->precision; 912 newPartsCount = partCountForBits(precision * 2); 913 914 if (newPartsCount > 4) 915 fullSignificand = new integerPart[newPartsCount]; 916 else 917 fullSignificand = scratch; 918 919 lhsSignificand = significandParts(); 920 partsCount = partCount(); 921 922 APInt::tcFullMultiply(fullSignificand, lhsSignificand, 923 rhs.significandParts(), partsCount, partsCount); 924 925 lost_fraction = lfExactlyZero; 926 omsb = APInt::tcMSB(fullSignificand, newPartsCount) + 1; 927 exponent += rhs.exponent; 928 929 // Assume the operands involved in the multiplication are single-precision 930 // FP, and the two multiplicants are: 931 // *this = a23 . a22 ... a0 * 2^e1 932 // rhs = b23 . b22 ... b0 * 2^e2 933 // the result of multiplication is: 934 // *this = c47 c46 . c45 ... c0 * 2^(e1+e2) 935 // Note that there are two significant bits at the left-hand side of the 936 // radix point. Move the radix point toward left by one bit, and adjust 937 // exponent accordingly. 938 exponent += 1; 939 940 if (addend) { 941 // The intermediate result of the multiplication has "2 * precision" 942 // signicant bit; adjust the addend to be consistent with mul result. 943 // 944 Significand savedSignificand = significand; 945 const fltSemantics *savedSemantics = semantics; 946 fltSemantics extendedSemantics; 947 opStatus status; 948 unsigned int extendedPrecision; 949 950 /* Normalize our MSB. */ 951 extendedPrecision = 2 * precision; 952 if (omsb != extendedPrecision) { 953 assert(extendedPrecision > omsb); 954 APInt::tcShiftLeft(fullSignificand, newPartsCount, 955 extendedPrecision - omsb); 956 exponent -= extendedPrecision - omsb; 957 } 958 959 /* Create new semantics. */ 960 extendedSemantics = *semantics; 961 extendedSemantics.precision = extendedPrecision; 962 963 if (newPartsCount == 1) 964 significand.part = fullSignificand[0]; 965 else 966 significand.parts = fullSignificand; 967 semantics = &extendedSemantics; 968 969 APFloat extendedAddend(*addend); 970 status = extendedAddend.convert(extendedSemantics, rmTowardZero, &ignored); 971 assert(status == opOK); 972 (void)status; 973 lost_fraction = addOrSubtractSignificand(extendedAddend, false); 974 975 /* Restore our state. */ 976 if (newPartsCount == 1) 977 fullSignificand[0] = significand.part; 978 significand = savedSignificand; 979 semantics = savedSemantics; 980 981 omsb = APInt::tcMSB(fullSignificand, newPartsCount) + 1; 982 } 983 984 // Convert the result having "2 * precision" significant-bits back to the one 985 // having "precision" significant-bits. First, move the radix point from 986 // poision "2*precision - 1" to "precision - 1". The exponent need to be 987 // adjusted by "2*precision - 1" - "precision - 1" = "precision". 988 exponent -= precision; 989 990 // In case MSB resides at the left-hand side of radix point, shift the 991 // mantissa right by some amount to make sure the MSB reside right before 992 // the radix point (i.e. "MSB . rest-significant-bits"). 993 // 994 // Note that the result is not normalized when "omsb < precision". So, the 995 // caller needs to call APFloat::normalize() if normalized value is expected. 996 if (omsb > precision) { 997 unsigned int bits, significantParts; 998 lostFraction lf; 999 1000 bits = omsb - precision; 1001 significantParts = partCountForBits(omsb); 1002 lf = shiftRight(fullSignificand, significantParts, bits); 1003 lost_fraction = combineLostFractions(lf, lost_fraction); 1004 exponent += bits; 1005 } 1006 1007 APInt::tcAssign(lhsSignificand, fullSignificand, partsCount); 1008 1009 if (newPartsCount > 4) 1010 delete [] fullSignificand; 1011 1012 return lost_fraction; 1013 } 1014 1015 /* Multiply the significands of LHS and RHS to DST. */ 1016 lostFraction 1017 APFloat::divideSignificand(const APFloat &rhs) 1018 { 1019 unsigned int bit, i, partsCount; 1020 const integerPart *rhsSignificand; 1021 integerPart *lhsSignificand, *dividend, *divisor; 1022 integerPart scratch[4]; 1023 lostFraction lost_fraction; 1024 1025 assert(semantics == rhs.semantics); 1026 1027 lhsSignificand = significandParts(); 1028 rhsSignificand = rhs.significandParts(); 1029 partsCount = partCount(); 1030 1031 if (partsCount > 2) 1032 dividend = new integerPart[partsCount * 2]; 1033 else 1034 dividend = scratch; 1035 1036 divisor = dividend + partsCount; 1037 1038 /* Copy the dividend and divisor as they will be modified in-place. */ 1039 for (i = 0; i < partsCount; i++) { 1040 dividend[i] = lhsSignificand[i]; 1041 divisor[i] = rhsSignificand[i]; 1042 lhsSignificand[i] = 0; 1043 } 1044 1045 exponent -= rhs.exponent; 1046 1047 unsigned int precision = semantics->precision; 1048 1049 /* Normalize the divisor. */ 1050 bit = precision - APInt::tcMSB(divisor, partsCount) - 1; 1051 if (bit) { 1052 exponent += bit; 1053 APInt::tcShiftLeft(divisor, partsCount, bit); 1054 } 1055 1056 /* Normalize the dividend. */ 1057 bit = precision - APInt::tcMSB(dividend, partsCount) - 1; 1058 if (bit) { 1059 exponent -= bit; 1060 APInt::tcShiftLeft(dividend, partsCount, bit); 1061 } 1062 1063 /* Ensure the dividend >= divisor initially for the loop below. 1064 Incidentally, this means that the division loop below is 1065 guaranteed to set the integer bit to one. */ 1066 if (APInt::tcCompare(dividend, divisor, partsCount) < 0) { 1067 exponent--; 1068 APInt::tcShiftLeft(dividend, partsCount, 1); 1069 assert(APInt::tcCompare(dividend, divisor, partsCount) >= 0); 1070 } 1071 1072 /* Long division. */ 1073 for (bit = precision; bit; bit -= 1) { 1074 if (APInt::tcCompare(dividend, divisor, partsCount) >= 0) { 1075 APInt::tcSubtract(dividend, divisor, 0, partsCount); 1076 APInt::tcSetBit(lhsSignificand, bit - 1); 1077 } 1078 1079 APInt::tcShiftLeft(dividend, partsCount, 1); 1080 } 1081 1082 /* Figure out the lost fraction. */ 1083 int cmp = APInt::tcCompare(dividend, divisor, partsCount); 1084 1085 if (cmp > 0) 1086 lost_fraction = lfMoreThanHalf; 1087 else if (cmp == 0) 1088 lost_fraction = lfExactlyHalf; 1089 else if (APInt::tcIsZero(dividend, partsCount)) 1090 lost_fraction = lfExactlyZero; 1091 else 1092 lost_fraction = lfLessThanHalf; 1093 1094 if (partsCount > 2) 1095 delete [] dividend; 1096 1097 return lost_fraction; 1098 } 1099 1100 unsigned int 1101 APFloat::significandMSB() const 1102 { 1103 return APInt::tcMSB(significandParts(), partCount()); 1104 } 1105 1106 unsigned int 1107 APFloat::significandLSB() const 1108 { 1109 return APInt::tcLSB(significandParts(), partCount()); 1110 } 1111 1112 /* Note that a zero result is NOT normalized to fcZero. */ 1113 lostFraction 1114 APFloat::shiftSignificandRight(unsigned int bits) 1115 { 1116 /* Our exponent should not overflow. */ 1117 assert((ExponentType) (exponent + bits) >= exponent); 1118 1119 exponent += bits; 1120 1121 return shiftRight(significandParts(), partCount(), bits); 1122 } 1123 1124 /* Shift the significand left BITS bits, subtract BITS from its exponent. */ 1125 void 1126 APFloat::shiftSignificandLeft(unsigned int bits) 1127 { 1128 assert(bits < semantics->precision); 1129 1130 if (bits) { 1131 unsigned int partsCount = partCount(); 1132 1133 APInt::tcShiftLeft(significandParts(), partsCount, bits); 1134 exponent -= bits; 1135 1136 assert(!APInt::tcIsZero(significandParts(), partsCount)); 1137 } 1138 } 1139 1140 APFloat::cmpResult 1141 APFloat::compareAbsoluteValue(const APFloat &rhs) const 1142 { 1143 int compare; 1144 1145 assert(semantics == rhs.semantics); 1146 assert(isFiniteNonZero()); 1147 assert(rhs.isFiniteNonZero()); 1148 1149 compare = exponent - rhs.exponent; 1150 1151 /* If exponents are equal, do an unsigned bignum comparison of the 1152 significands. */ 1153 if (compare == 0) 1154 compare = APInt::tcCompare(significandParts(), rhs.significandParts(), 1155 partCount()); 1156 1157 if (compare > 0) 1158 return cmpGreaterThan; 1159 else if (compare < 0) 1160 return cmpLessThan; 1161 else 1162 return cmpEqual; 1163 } 1164 1165 /* Handle overflow. Sign is preserved. We either become infinity or 1166 the largest finite number. */ 1167 APFloat::opStatus 1168 APFloat::handleOverflow(roundingMode rounding_mode) 1169 { 1170 /* Infinity? */ 1171 if (rounding_mode == rmNearestTiesToEven || 1172 rounding_mode == rmNearestTiesToAway || 1173 (rounding_mode == rmTowardPositive && !sign) || 1174 (rounding_mode == rmTowardNegative && sign)) { 1175 category = fcInfinity; 1176 return (opStatus) (opOverflow | opInexact); 1177 } 1178 1179 /* Otherwise we become the largest finite number. */ 1180 category = fcNormal; 1181 exponent = semantics->maxExponent; 1182 APInt::tcSetLeastSignificantBits(significandParts(), partCount(), 1183 semantics->precision); 1184 1185 return opInexact; 1186 } 1187 1188 /* Returns TRUE if, when truncating the current number, with BIT the 1189 new LSB, with the given lost fraction and rounding mode, the result 1190 would need to be rounded away from zero (i.e., by increasing the 1191 signficand). This routine must work for fcZero of both signs, and 1192 fcNormal numbers. */ 1193 bool 1194 APFloat::roundAwayFromZero(roundingMode rounding_mode, 1195 lostFraction lost_fraction, 1196 unsigned int bit) const 1197 { 1198 /* NaNs and infinities should not have lost fractions. */ 1199 assert(isFiniteNonZero() || category == fcZero); 1200 1201 /* Current callers never pass this so we don't handle it. */ 1202 assert(lost_fraction != lfExactlyZero); 1203 1204 switch (rounding_mode) { 1205 case rmNearestTiesToAway: 1206 return lost_fraction == lfExactlyHalf || lost_fraction == lfMoreThanHalf; 1207 1208 case rmNearestTiesToEven: 1209 if (lost_fraction == lfMoreThanHalf) 1210 return true; 1211 1212 /* Our zeroes don't have a significand to test. */ 1213 if (lost_fraction == lfExactlyHalf && category != fcZero) 1214 return APInt::tcExtractBit(significandParts(), bit); 1215 1216 return false; 1217 1218 case rmTowardZero: 1219 return false; 1220 1221 case rmTowardPositive: 1222 return sign == false; 1223 1224 case rmTowardNegative: 1225 return sign == true; 1226 } 1227 llvm_unreachable("Invalid rounding mode found"); 1228 } 1229 1230 APFloat::opStatus 1231 APFloat::normalize(roundingMode rounding_mode, 1232 lostFraction lost_fraction) 1233 { 1234 unsigned int omsb; /* One, not zero, based MSB. */ 1235 int exponentChange; 1236 1237 if (!isFiniteNonZero()) 1238 return opOK; 1239 1240 /* Before rounding normalize the exponent of fcNormal numbers. */ 1241 omsb = significandMSB() + 1; 1242 1243 if (omsb) { 1244 /* OMSB is numbered from 1. We want to place it in the integer 1245 bit numbered PRECISION if possible, with a compensating change in 1246 the exponent. */ 1247 exponentChange = omsb - semantics->precision; 1248 1249 /* If the resulting exponent is too high, overflow according to 1250 the rounding mode. */ 1251 if (exponent + exponentChange > semantics->maxExponent) 1252 return handleOverflow(rounding_mode); 1253 1254 /* Subnormal numbers have exponent minExponent, and their MSB 1255 is forced based on that. */ 1256 if (exponent + exponentChange < semantics->minExponent) 1257 exponentChange = semantics->minExponent - exponent; 1258 1259 /* Shifting left is easy as we don't lose precision. */ 1260 if (exponentChange < 0) { 1261 assert(lost_fraction == lfExactlyZero); 1262 1263 shiftSignificandLeft(-exponentChange); 1264 1265 return opOK; 1266 } 1267 1268 if (exponentChange > 0) { 1269 lostFraction lf; 1270 1271 /* Shift right and capture any new lost fraction. */ 1272 lf = shiftSignificandRight(exponentChange); 1273 1274 lost_fraction = combineLostFractions(lf, lost_fraction); 1275 1276 /* Keep OMSB up-to-date. */ 1277 if (omsb > (unsigned) exponentChange) 1278 omsb -= exponentChange; 1279 else 1280 omsb = 0; 1281 } 1282 } 1283 1284 /* Now round the number according to rounding_mode given the lost 1285 fraction. */ 1286 1287 /* As specified in IEEE 754, since we do not trap we do not report 1288 underflow for exact results. */ 1289 if (lost_fraction == lfExactlyZero) { 1290 /* Canonicalize zeroes. */ 1291 if (omsb == 0) 1292 category = fcZero; 1293 1294 return opOK; 1295 } 1296 1297 /* Increment the significand if we're rounding away from zero. */ 1298 if (roundAwayFromZero(rounding_mode, lost_fraction, 0)) { 1299 if (omsb == 0) 1300 exponent = semantics->minExponent; 1301 1302 incrementSignificand(); 1303 omsb = significandMSB() + 1; 1304 1305 /* Did the significand increment overflow? */ 1306 if (omsb == (unsigned) semantics->precision + 1) { 1307 /* Renormalize by incrementing the exponent and shifting our 1308 significand right one. However if we already have the 1309 maximum exponent we overflow to infinity. */ 1310 if (exponent == semantics->maxExponent) { 1311 category = fcInfinity; 1312 1313 return (opStatus) (opOverflow | opInexact); 1314 } 1315 1316 shiftSignificandRight(1); 1317 1318 return opInexact; 1319 } 1320 } 1321 1322 /* The normal case - we were and are not denormal, and any 1323 significand increment above didn't overflow. */ 1324 if (omsb == semantics->precision) 1325 return opInexact; 1326 1327 /* We have a non-zero denormal. */ 1328 assert(omsb < semantics->precision); 1329 1330 /* Canonicalize zeroes. */ 1331 if (omsb == 0) 1332 category = fcZero; 1333 1334 /* The fcZero case is a denormal that underflowed to zero. */ 1335 return (opStatus) (opUnderflow | opInexact); 1336 } 1337 1338 APFloat::opStatus 1339 APFloat::addOrSubtractSpecials(const APFloat &rhs, bool subtract) 1340 { 1341 switch (PackCategoriesIntoKey(category, rhs.category)) { 1342 default: 1343 llvm_unreachable(0); 1344 1345 case PackCategoriesIntoKey(fcNaN, fcZero): 1346 case PackCategoriesIntoKey(fcNaN, fcNormal): 1347 case PackCategoriesIntoKey(fcNaN, fcInfinity): 1348 case PackCategoriesIntoKey(fcNaN, fcNaN): 1349 case PackCategoriesIntoKey(fcNormal, fcZero): 1350 case PackCategoriesIntoKey(fcInfinity, fcNormal): 1351 case PackCategoriesIntoKey(fcInfinity, fcZero): 1352 return opOK; 1353 1354 case PackCategoriesIntoKey(fcZero, fcNaN): 1355 case PackCategoriesIntoKey(fcNormal, fcNaN): 1356 case PackCategoriesIntoKey(fcInfinity, fcNaN): 1357 category = fcNaN; 1358 copySignificand(rhs); 1359 return opOK; 1360 1361 case PackCategoriesIntoKey(fcNormal, fcInfinity): 1362 case PackCategoriesIntoKey(fcZero, fcInfinity): 1363 category = fcInfinity; 1364 sign = rhs.sign ^ subtract; 1365 return opOK; 1366 1367 case PackCategoriesIntoKey(fcZero, fcNormal): 1368 assign(rhs); 1369 sign = rhs.sign ^ subtract; 1370 return opOK; 1371 1372 case PackCategoriesIntoKey(fcZero, fcZero): 1373 /* Sign depends on rounding mode; handled by caller. */ 1374 return opOK; 1375 1376 case PackCategoriesIntoKey(fcInfinity, fcInfinity): 1377 /* Differently signed infinities can only be validly 1378 subtracted. */ 1379 if (((sign ^ rhs.sign)!=0) != subtract) { 1380 makeNaN(); 1381 return opInvalidOp; 1382 } 1383 1384 return opOK; 1385 1386 case PackCategoriesIntoKey(fcNormal, fcNormal): 1387 return opDivByZero; 1388 } 1389 } 1390 1391 /* Add or subtract two normal numbers. */ 1392 lostFraction 1393 APFloat::addOrSubtractSignificand(const APFloat &rhs, bool subtract) 1394 { 1395 integerPart carry; 1396 lostFraction lost_fraction; 1397 int bits; 1398 1399 /* Determine if the operation on the absolute values is effectively 1400 an addition or subtraction. */ 1401 subtract ^= (sign ^ rhs.sign) ? true : false; 1402 1403 /* Are we bigger exponent-wise than the RHS? */ 1404 bits = exponent - rhs.exponent; 1405 1406 /* Subtraction is more subtle than one might naively expect. */ 1407 if (subtract) { 1408 APFloat temp_rhs(rhs); 1409 bool reverse; 1410 1411 if (bits == 0) { 1412 reverse = compareAbsoluteValue(temp_rhs) == cmpLessThan; 1413 lost_fraction = lfExactlyZero; 1414 } else if (bits > 0) { 1415 lost_fraction = temp_rhs.shiftSignificandRight(bits - 1); 1416 shiftSignificandLeft(1); 1417 reverse = false; 1418 } else { 1419 lost_fraction = shiftSignificandRight(-bits - 1); 1420 temp_rhs.shiftSignificandLeft(1); 1421 reverse = true; 1422 } 1423 1424 if (reverse) { 1425 carry = temp_rhs.subtractSignificand 1426 (*this, lost_fraction != lfExactlyZero); 1427 copySignificand(temp_rhs); 1428 sign = !sign; 1429 } else { 1430 carry = subtractSignificand 1431 (temp_rhs, lost_fraction != lfExactlyZero); 1432 } 1433 1434 /* Invert the lost fraction - it was on the RHS and 1435 subtracted. */ 1436 if (lost_fraction == lfLessThanHalf) 1437 lost_fraction = lfMoreThanHalf; 1438 else if (lost_fraction == lfMoreThanHalf) 1439 lost_fraction = lfLessThanHalf; 1440 1441 /* The code above is intended to ensure that no borrow is 1442 necessary. */ 1443 assert(!carry); 1444 (void)carry; 1445 } else { 1446 if (bits > 0) { 1447 APFloat temp_rhs(rhs); 1448 1449 lost_fraction = temp_rhs.shiftSignificandRight(bits); 1450 carry = addSignificand(temp_rhs); 1451 } else { 1452 lost_fraction = shiftSignificandRight(-bits); 1453 carry = addSignificand(rhs); 1454 } 1455 1456 /* We have a guard bit; generating a carry cannot happen. */ 1457 assert(!carry); 1458 (void)carry; 1459 } 1460 1461 return lost_fraction; 1462 } 1463 1464 APFloat::opStatus 1465 APFloat::multiplySpecials(const APFloat &rhs) 1466 { 1467 switch (PackCategoriesIntoKey(category, rhs.category)) { 1468 default: 1469 llvm_unreachable(0); 1470 1471 case PackCategoriesIntoKey(fcNaN, fcZero): 1472 case PackCategoriesIntoKey(fcNaN, fcNormal): 1473 case PackCategoriesIntoKey(fcNaN, fcInfinity): 1474 case PackCategoriesIntoKey(fcNaN, fcNaN): 1475 return opOK; 1476 1477 case PackCategoriesIntoKey(fcZero, fcNaN): 1478 case PackCategoriesIntoKey(fcNormal, fcNaN): 1479 case PackCategoriesIntoKey(fcInfinity, fcNaN): 1480 category = fcNaN; 1481 copySignificand(rhs); 1482 return opOK; 1483 1484 case PackCategoriesIntoKey(fcNormal, fcInfinity): 1485 case PackCategoriesIntoKey(fcInfinity, fcNormal): 1486 case PackCategoriesIntoKey(fcInfinity, fcInfinity): 1487 category = fcInfinity; 1488 return opOK; 1489 1490 case PackCategoriesIntoKey(fcZero, fcNormal): 1491 case PackCategoriesIntoKey(fcNormal, fcZero): 1492 case PackCategoriesIntoKey(fcZero, fcZero): 1493 category = fcZero; 1494 return opOK; 1495 1496 case PackCategoriesIntoKey(fcZero, fcInfinity): 1497 case PackCategoriesIntoKey(fcInfinity, fcZero): 1498 makeNaN(); 1499 return opInvalidOp; 1500 1501 case PackCategoriesIntoKey(fcNormal, fcNormal): 1502 return opOK; 1503 } 1504 } 1505 1506 APFloat::opStatus 1507 APFloat::divideSpecials(const APFloat &rhs) 1508 { 1509 switch (PackCategoriesIntoKey(category, rhs.category)) { 1510 default: 1511 llvm_unreachable(0); 1512 1513 case PackCategoriesIntoKey(fcNaN, fcZero): 1514 case PackCategoriesIntoKey(fcNaN, fcNormal): 1515 case PackCategoriesIntoKey(fcNaN, fcInfinity): 1516 case PackCategoriesIntoKey(fcNaN, fcNaN): 1517 case PackCategoriesIntoKey(fcInfinity, fcZero): 1518 case PackCategoriesIntoKey(fcInfinity, fcNormal): 1519 case PackCategoriesIntoKey(fcZero, fcInfinity): 1520 case PackCategoriesIntoKey(fcZero, fcNormal): 1521 return opOK; 1522 1523 case PackCategoriesIntoKey(fcZero, fcNaN): 1524 case PackCategoriesIntoKey(fcNormal, fcNaN): 1525 case PackCategoriesIntoKey(fcInfinity, fcNaN): 1526 category = fcNaN; 1527 copySignificand(rhs); 1528 return opOK; 1529 1530 case PackCategoriesIntoKey(fcNormal, fcInfinity): 1531 category = fcZero; 1532 return opOK; 1533 1534 case PackCategoriesIntoKey(fcNormal, fcZero): 1535 category = fcInfinity; 1536 return opDivByZero; 1537 1538 case PackCategoriesIntoKey(fcInfinity, fcInfinity): 1539 case PackCategoriesIntoKey(fcZero, fcZero): 1540 makeNaN(); 1541 return opInvalidOp; 1542 1543 case PackCategoriesIntoKey(fcNormal, fcNormal): 1544 return opOK; 1545 } 1546 } 1547 1548 APFloat::opStatus 1549 APFloat::modSpecials(const APFloat &rhs) 1550 { 1551 switch (PackCategoriesIntoKey(category, rhs.category)) { 1552 default: 1553 llvm_unreachable(0); 1554 1555 case PackCategoriesIntoKey(fcNaN, fcZero): 1556 case PackCategoriesIntoKey(fcNaN, fcNormal): 1557 case PackCategoriesIntoKey(fcNaN, fcInfinity): 1558 case PackCategoriesIntoKey(fcNaN, fcNaN): 1559 case PackCategoriesIntoKey(fcZero, fcInfinity): 1560 case PackCategoriesIntoKey(fcZero, fcNormal): 1561 case PackCategoriesIntoKey(fcNormal, fcInfinity): 1562 return opOK; 1563 1564 case PackCategoriesIntoKey(fcZero, fcNaN): 1565 case PackCategoriesIntoKey(fcNormal, fcNaN): 1566 case PackCategoriesIntoKey(fcInfinity, fcNaN): 1567 category = fcNaN; 1568 copySignificand(rhs); 1569 return opOK; 1570 1571 case PackCategoriesIntoKey(fcNormal, fcZero): 1572 case PackCategoriesIntoKey(fcInfinity, fcZero): 1573 case PackCategoriesIntoKey(fcInfinity, fcNormal): 1574 case PackCategoriesIntoKey(fcInfinity, fcInfinity): 1575 case PackCategoriesIntoKey(fcZero, fcZero): 1576 makeNaN(); 1577 return opInvalidOp; 1578 1579 case PackCategoriesIntoKey(fcNormal, fcNormal): 1580 return opOK; 1581 } 1582 } 1583 1584 /* Change sign. */ 1585 void 1586 APFloat::changeSign() 1587 { 1588 /* Look mummy, this one's easy. */ 1589 sign = !sign; 1590 } 1591 1592 void 1593 APFloat::clearSign() 1594 { 1595 /* So is this one. */ 1596 sign = 0; 1597 } 1598 1599 void 1600 APFloat::copySign(const APFloat &rhs) 1601 { 1602 /* And this one. */ 1603 sign = rhs.sign; 1604 } 1605 1606 /* Normalized addition or subtraction. */ 1607 APFloat::opStatus 1608 APFloat::addOrSubtract(const APFloat &rhs, roundingMode rounding_mode, 1609 bool subtract) 1610 { 1611 opStatus fs; 1612 1613 fs = addOrSubtractSpecials(rhs, subtract); 1614 1615 /* This return code means it was not a simple case. */ 1616 if (fs == opDivByZero) { 1617 lostFraction lost_fraction; 1618 1619 lost_fraction = addOrSubtractSignificand(rhs, subtract); 1620 fs = normalize(rounding_mode, lost_fraction); 1621 1622 /* Can only be zero if we lost no fraction. */ 1623 assert(category != fcZero || lost_fraction == lfExactlyZero); 1624 } 1625 1626 /* If two numbers add (exactly) to zero, IEEE 754 decrees it is a 1627 positive zero unless rounding to minus infinity, except that 1628 adding two like-signed zeroes gives that zero. */ 1629 if (category == fcZero) { 1630 if (rhs.category != fcZero || (sign == rhs.sign) == subtract) 1631 sign = (rounding_mode == rmTowardNegative); 1632 } 1633 1634 return fs; 1635 } 1636 1637 /* Normalized addition. */ 1638 APFloat::opStatus 1639 APFloat::add(const APFloat &rhs, roundingMode rounding_mode) 1640 { 1641 return addOrSubtract(rhs, rounding_mode, false); 1642 } 1643 1644 /* Normalized subtraction. */ 1645 APFloat::opStatus 1646 APFloat::subtract(const APFloat &rhs, roundingMode rounding_mode) 1647 { 1648 return addOrSubtract(rhs, rounding_mode, true); 1649 } 1650 1651 /* Normalized multiply. */ 1652 APFloat::opStatus 1653 APFloat::multiply(const APFloat &rhs, roundingMode rounding_mode) 1654 { 1655 opStatus fs; 1656 1657 sign ^= rhs.sign; 1658 fs = multiplySpecials(rhs); 1659 1660 if (isFiniteNonZero()) { 1661 lostFraction lost_fraction = multiplySignificand(rhs, 0); 1662 fs = normalize(rounding_mode, lost_fraction); 1663 if (lost_fraction != lfExactlyZero) 1664 fs = (opStatus) (fs | opInexact); 1665 } 1666 1667 return fs; 1668 } 1669 1670 /* Normalized divide. */ 1671 APFloat::opStatus 1672 APFloat::divide(const APFloat &rhs, roundingMode rounding_mode) 1673 { 1674 opStatus fs; 1675 1676 sign ^= rhs.sign; 1677 fs = divideSpecials(rhs); 1678 1679 if (isFiniteNonZero()) { 1680 lostFraction lost_fraction = divideSignificand(rhs); 1681 fs = normalize(rounding_mode, lost_fraction); 1682 if (lost_fraction != lfExactlyZero) 1683 fs = (opStatus) (fs | opInexact); 1684 } 1685 1686 return fs; 1687 } 1688 1689 /* Normalized remainder. This is not currently correct in all cases. */ 1690 APFloat::opStatus 1691 APFloat::remainder(const APFloat &rhs) 1692 { 1693 opStatus fs; 1694 APFloat V = *this; 1695 unsigned int origSign = sign; 1696 1697 fs = V.divide(rhs, rmNearestTiesToEven); 1698 if (fs == opDivByZero) 1699 return fs; 1700 1701 int parts = partCount(); 1702 integerPart *x = new integerPart[parts]; 1703 bool ignored; 1704 fs = V.convertToInteger(x, parts * integerPartWidth, true, 1705 rmNearestTiesToEven, &ignored); 1706 if (fs==opInvalidOp) 1707 return fs; 1708 1709 fs = V.convertFromZeroExtendedInteger(x, parts * integerPartWidth, true, 1710 rmNearestTiesToEven); 1711 assert(fs==opOK); // should always work 1712 1713 fs = V.multiply(rhs, rmNearestTiesToEven); 1714 assert(fs==opOK || fs==opInexact); // should not overflow or underflow 1715 1716 fs = subtract(V, rmNearestTiesToEven); 1717 assert(fs==opOK || fs==opInexact); // likewise 1718 1719 if (isZero()) 1720 sign = origSign; // IEEE754 requires this 1721 delete[] x; 1722 return fs; 1723 } 1724 1725 /* Normalized llvm frem (C fmod). 1726 This is not currently correct in all cases. */ 1727 APFloat::opStatus 1728 APFloat::mod(const APFloat &rhs, roundingMode rounding_mode) 1729 { 1730 opStatus fs; 1731 fs = modSpecials(rhs); 1732 1733 if (isFiniteNonZero() && rhs.isFiniteNonZero()) { 1734 APFloat V = *this; 1735 unsigned int origSign = sign; 1736 1737 fs = V.divide(rhs, rmNearestTiesToEven); 1738 if (fs == opDivByZero) 1739 return fs; 1740 1741 int parts = partCount(); 1742 integerPart *x = new integerPart[parts]; 1743 bool ignored; 1744 fs = V.convertToInteger(x, parts * integerPartWidth, true, 1745 rmTowardZero, &ignored); 1746 if (fs==opInvalidOp) 1747 return fs; 1748 1749 fs = V.convertFromZeroExtendedInteger(x, parts * integerPartWidth, true, 1750 rmNearestTiesToEven); 1751 assert(fs==opOK); // should always work 1752 1753 fs = V.multiply(rhs, rounding_mode); 1754 assert(fs==opOK || fs==opInexact); // should not overflow or underflow 1755 1756 fs = subtract(V, rounding_mode); 1757 assert(fs==opOK || fs==opInexact); // likewise 1758 1759 if (isZero()) 1760 sign = origSign; // IEEE754 requires this 1761 delete[] x; 1762 } 1763 return fs; 1764 } 1765 1766 /* Normalized fused-multiply-add. */ 1767 APFloat::opStatus 1768 APFloat::fusedMultiplyAdd(const APFloat &multiplicand, 1769 const APFloat &addend, 1770 roundingMode rounding_mode) 1771 { 1772 opStatus fs; 1773 1774 /* Post-multiplication sign, before addition. */ 1775 sign ^= multiplicand.sign; 1776 1777 /* If and only if all arguments are normal do we need to do an 1778 extended-precision calculation. */ 1779 if (isFiniteNonZero() && 1780 multiplicand.isFiniteNonZero() && 1781 addend.isFiniteNonZero()) { 1782 lostFraction lost_fraction; 1783 1784 lost_fraction = multiplySignificand(multiplicand, &addend); 1785 fs = normalize(rounding_mode, lost_fraction); 1786 if (lost_fraction != lfExactlyZero) 1787 fs = (opStatus) (fs | opInexact); 1788 1789 /* If two numbers add (exactly) to zero, IEEE 754 decrees it is a 1790 positive zero unless rounding to minus infinity, except that 1791 adding two like-signed zeroes gives that zero. */ 1792 if (category == fcZero && sign != addend.sign) 1793 sign = (rounding_mode == rmTowardNegative); 1794 } else { 1795 fs = multiplySpecials(multiplicand); 1796 1797 /* FS can only be opOK or opInvalidOp. There is no more work 1798 to do in the latter case. The IEEE-754R standard says it is 1799 implementation-defined in this case whether, if ADDEND is a 1800 quiet NaN, we raise invalid op; this implementation does so. 1801 1802 If we need to do the addition we can do so with normal 1803 precision. */ 1804 if (fs == opOK) 1805 fs = addOrSubtract(addend, rounding_mode, false); 1806 } 1807 1808 return fs; 1809 } 1810 1811 /* Rounding-mode corrrect round to integral value. */ 1812 APFloat::opStatus APFloat::roundToIntegral(roundingMode rounding_mode) { 1813 opStatus fs; 1814 1815 // If the exponent is large enough, we know that this value is already 1816 // integral, and the arithmetic below would potentially cause it to saturate 1817 // to +/-Inf. Bail out early instead. 1818 if (isFiniteNonZero() && exponent+1 >= (int)semanticsPrecision(*semantics)) 1819 return opOK; 1820 1821 // The algorithm here is quite simple: we add 2^(p-1), where p is the 1822 // precision of our format, and then subtract it back off again. The choice 1823 // of rounding modes for the addition/subtraction determines the rounding mode 1824 // for our integral rounding as well. 1825 // NOTE: When the input value is negative, we do subtraction followed by 1826 // addition instead. 1827 APInt IntegerConstant(NextPowerOf2(semanticsPrecision(*semantics)), 1); 1828 IntegerConstant <<= semanticsPrecision(*semantics)-1; 1829 APFloat MagicConstant(*semantics); 1830 fs = MagicConstant.convertFromAPInt(IntegerConstant, false, 1831 rmNearestTiesToEven); 1832 MagicConstant.copySign(*this); 1833 1834 if (fs != opOK) 1835 return fs; 1836 1837 // Preserve the input sign so that we can handle 0.0/-0.0 cases correctly. 1838 bool inputSign = isNegative(); 1839 1840 fs = add(MagicConstant, rounding_mode); 1841 if (fs != opOK && fs != opInexact) 1842 return fs; 1843 1844 fs = subtract(MagicConstant, rounding_mode); 1845 1846 // Restore the input sign. 1847 if (inputSign != isNegative()) 1848 changeSign(); 1849 1850 return fs; 1851 } 1852 1853 1854 /* Comparison requires normalized numbers. */ 1855 APFloat::cmpResult 1856 APFloat::compare(const APFloat &rhs) const 1857 { 1858 cmpResult result; 1859 1860 assert(semantics == rhs.semantics); 1861 1862 switch (PackCategoriesIntoKey(category, rhs.category)) { 1863 default: 1864 llvm_unreachable(0); 1865 1866 case PackCategoriesIntoKey(fcNaN, fcZero): 1867 case PackCategoriesIntoKey(fcNaN, fcNormal): 1868 case PackCategoriesIntoKey(fcNaN, fcInfinity): 1869 case PackCategoriesIntoKey(fcNaN, fcNaN): 1870 case PackCategoriesIntoKey(fcZero, fcNaN): 1871 case PackCategoriesIntoKey(fcNormal, fcNaN): 1872 case PackCategoriesIntoKey(fcInfinity, fcNaN): 1873 return cmpUnordered; 1874 1875 case PackCategoriesIntoKey(fcInfinity, fcNormal): 1876 case PackCategoriesIntoKey(fcInfinity, fcZero): 1877 case PackCategoriesIntoKey(fcNormal, fcZero): 1878 if (sign) 1879 return cmpLessThan; 1880 else 1881 return cmpGreaterThan; 1882 1883 case PackCategoriesIntoKey(fcNormal, fcInfinity): 1884 case PackCategoriesIntoKey(fcZero, fcInfinity): 1885 case PackCategoriesIntoKey(fcZero, fcNormal): 1886 if (rhs.sign) 1887 return cmpGreaterThan; 1888 else 1889 return cmpLessThan; 1890 1891 case PackCategoriesIntoKey(fcInfinity, fcInfinity): 1892 if (sign == rhs.sign) 1893 return cmpEqual; 1894 else if (sign) 1895 return cmpLessThan; 1896 else 1897 return cmpGreaterThan; 1898 1899 case PackCategoriesIntoKey(fcZero, fcZero): 1900 return cmpEqual; 1901 1902 case PackCategoriesIntoKey(fcNormal, fcNormal): 1903 break; 1904 } 1905 1906 /* Two normal numbers. Do they have the same sign? */ 1907 if (sign != rhs.sign) { 1908 if (sign) 1909 result = cmpLessThan; 1910 else 1911 result = cmpGreaterThan; 1912 } else { 1913 /* Compare absolute values; invert result if negative. */ 1914 result = compareAbsoluteValue(rhs); 1915 1916 if (sign) { 1917 if (result == cmpLessThan) 1918 result = cmpGreaterThan; 1919 else if (result == cmpGreaterThan) 1920 result = cmpLessThan; 1921 } 1922 } 1923 1924 return result; 1925 } 1926 1927 /// APFloat::convert - convert a value of one floating point type to another. 1928 /// The return value corresponds to the IEEE754 exceptions. *losesInfo 1929 /// records whether the transformation lost information, i.e. whether 1930 /// converting the result back to the original type will produce the 1931 /// original value (this is almost the same as return value==fsOK, but there 1932 /// are edge cases where this is not so). 1933 1934 APFloat::opStatus 1935 APFloat::convert(const fltSemantics &toSemantics, 1936 roundingMode rounding_mode, bool *losesInfo) 1937 { 1938 lostFraction lostFraction; 1939 unsigned int newPartCount, oldPartCount; 1940 opStatus fs; 1941 int shift; 1942 const fltSemantics &fromSemantics = *semantics; 1943 1944 lostFraction = lfExactlyZero; 1945 newPartCount = partCountForBits(toSemantics.precision + 1); 1946 oldPartCount = partCount(); 1947 shift = toSemantics.precision - fromSemantics.precision; 1948 1949 bool X86SpecialNan = false; 1950 if (&fromSemantics == &APFloat::x87DoubleExtended && 1951 &toSemantics != &APFloat::x87DoubleExtended && category == fcNaN && 1952 (!(*significandParts() & 0x8000000000000000ULL) || 1953 !(*significandParts() & 0x4000000000000000ULL))) { 1954 // x86 has some unusual NaNs which cannot be represented in any other 1955 // format; note them here. 1956 X86SpecialNan = true; 1957 } 1958 1959 // If this is a truncation of a denormal number, and the target semantics 1960 // has larger exponent range than the source semantics (this can happen 1961 // when truncating from PowerPC double-double to double format), the 1962 // right shift could lose result mantissa bits. Adjust exponent instead 1963 // of performing excessive shift. 1964 if (shift < 0 && isFiniteNonZero()) { 1965 int exponentChange = significandMSB() + 1 - fromSemantics.precision; 1966 if (exponent + exponentChange < toSemantics.minExponent) 1967 exponentChange = toSemantics.minExponent - exponent; 1968 if (exponentChange < shift) 1969 exponentChange = shift; 1970 if (exponentChange < 0) { 1971 shift -= exponentChange; 1972 exponent += exponentChange; 1973 } 1974 } 1975 1976 // If this is a truncation, perform the shift before we narrow the storage. 1977 if (shift < 0 && (isFiniteNonZero() || category==fcNaN)) 1978 lostFraction = shiftRight(significandParts(), oldPartCount, -shift); 1979 1980 // Fix the storage so it can hold to new value. 1981 if (newPartCount > oldPartCount) { 1982 // The new type requires more storage; make it available. 1983 integerPart *newParts; 1984 newParts = new integerPart[newPartCount]; 1985 APInt::tcSet(newParts, 0, newPartCount); 1986 if (isFiniteNonZero() || category==fcNaN) 1987 APInt::tcAssign(newParts, significandParts(), oldPartCount); 1988 freeSignificand(); 1989 significand.parts = newParts; 1990 } else if (newPartCount == 1 && oldPartCount != 1) { 1991 // Switch to built-in storage for a single part. 1992 integerPart newPart = 0; 1993 if (isFiniteNonZero() || category==fcNaN) 1994 newPart = significandParts()[0]; 1995 freeSignificand(); 1996 significand.part = newPart; 1997 } 1998 1999 // Now that we have the right storage, switch the semantics. 2000 semantics = &toSemantics; 2001 2002 // If this is an extension, perform the shift now that the storage is 2003 // available. 2004 if (shift > 0 && (isFiniteNonZero() || category==fcNaN)) 2005 APInt::tcShiftLeft(significandParts(), newPartCount, shift); 2006 2007 if (isFiniteNonZero()) { 2008 fs = normalize(rounding_mode, lostFraction); 2009 *losesInfo = (fs != opOK); 2010 } else if (category == fcNaN) { 2011 *losesInfo = lostFraction != lfExactlyZero || X86SpecialNan; 2012 2013 // For x87 extended precision, we want to make a NaN, not a special NaN if 2014 // the input wasn't special either. 2015 if (!X86SpecialNan && semantics == &APFloat::x87DoubleExtended) 2016 APInt::tcSetBit(significandParts(), semantics->precision - 1); 2017 2018 // gcc forces the Quiet bit on, which means (float)(double)(float_sNan) 2019 // does not give you back the same bits. This is dubious, and we 2020 // don't currently do it. You're really supposed to get 2021 // an invalid operation signal at runtime, but nobody does that. 2022 fs = opOK; 2023 } else { 2024 *losesInfo = false; 2025 fs = opOK; 2026 } 2027 2028 return fs; 2029 } 2030 2031 /* Convert a floating point number to an integer according to the 2032 rounding mode. If the rounded integer value is out of range this 2033 returns an invalid operation exception and the contents of the 2034 destination parts are unspecified. If the rounded value is in 2035 range but the floating point number is not the exact integer, the C 2036 standard doesn't require an inexact exception to be raised. IEEE 2037 854 does require it so we do that. 2038 2039 Note that for conversions to integer type the C standard requires 2040 round-to-zero to always be used. */ 2041 APFloat::opStatus 2042 APFloat::convertToSignExtendedInteger(integerPart *parts, unsigned int width, 2043 bool isSigned, 2044 roundingMode rounding_mode, 2045 bool *isExact) const 2046 { 2047 lostFraction lost_fraction; 2048 const integerPart *src; 2049 unsigned int dstPartsCount, truncatedBits; 2050 2051 *isExact = false; 2052 2053 /* Handle the three special cases first. */ 2054 if (category == fcInfinity || category == fcNaN) 2055 return opInvalidOp; 2056 2057 dstPartsCount = partCountForBits(width); 2058 2059 if (category == fcZero) { 2060 APInt::tcSet(parts, 0, dstPartsCount); 2061 // Negative zero can't be represented as an int. 2062 *isExact = !sign; 2063 return opOK; 2064 } 2065 2066 src = significandParts(); 2067 2068 /* Step 1: place our absolute value, with any fraction truncated, in 2069 the destination. */ 2070 if (exponent < 0) { 2071 /* Our absolute value is less than one; truncate everything. */ 2072 APInt::tcSet(parts, 0, dstPartsCount); 2073 /* For exponent -1 the integer bit represents .5, look at that. 2074 For smaller exponents leftmost truncated bit is 0. */ 2075 truncatedBits = semantics->precision -1U - exponent; 2076 } else { 2077 /* We want the most significant (exponent + 1) bits; the rest are 2078 truncated. */ 2079 unsigned int bits = exponent + 1U; 2080 2081 /* Hopelessly large in magnitude? */ 2082 if (bits > width) 2083 return opInvalidOp; 2084 2085 if (bits < semantics->precision) { 2086 /* We truncate (semantics->precision - bits) bits. */ 2087 truncatedBits = semantics->precision - bits; 2088 APInt::tcExtract(parts, dstPartsCount, src, bits, truncatedBits); 2089 } else { 2090 /* We want at least as many bits as are available. */ 2091 APInt::tcExtract(parts, dstPartsCount, src, semantics->precision, 0); 2092 APInt::tcShiftLeft(parts, dstPartsCount, bits - semantics->precision); 2093 truncatedBits = 0; 2094 } 2095 } 2096 2097 /* Step 2: work out any lost fraction, and increment the absolute 2098 value if we would round away from zero. */ 2099 if (truncatedBits) { 2100 lost_fraction = lostFractionThroughTruncation(src, partCount(), 2101 truncatedBits); 2102 if (lost_fraction != lfExactlyZero && 2103 roundAwayFromZero(rounding_mode, lost_fraction, truncatedBits)) { 2104 if (APInt::tcIncrement(parts, dstPartsCount)) 2105 return opInvalidOp; /* Overflow. */ 2106 } 2107 } else { 2108 lost_fraction = lfExactlyZero; 2109 } 2110 2111 /* Step 3: check if we fit in the destination. */ 2112 unsigned int omsb = APInt::tcMSB(parts, dstPartsCount) + 1; 2113 2114 if (sign) { 2115 if (!isSigned) { 2116 /* Negative numbers cannot be represented as unsigned. */ 2117 if (omsb != 0) 2118 return opInvalidOp; 2119 } else { 2120 /* It takes omsb bits to represent the unsigned integer value. 2121 We lose a bit for the sign, but care is needed as the 2122 maximally negative integer is a special case. */ 2123 if (omsb == width && APInt::tcLSB(parts, dstPartsCount) + 1 != omsb) 2124 return opInvalidOp; 2125 2126 /* This case can happen because of rounding. */ 2127 if (omsb > width) 2128 return opInvalidOp; 2129 } 2130 2131 APInt::tcNegate (parts, dstPartsCount); 2132 } else { 2133 if (omsb >= width + !isSigned) 2134 return opInvalidOp; 2135 } 2136 2137 if (lost_fraction == lfExactlyZero) { 2138 *isExact = true; 2139 return opOK; 2140 } else 2141 return opInexact; 2142 } 2143 2144 /* Same as convertToSignExtendedInteger, except we provide 2145 deterministic values in case of an invalid operation exception, 2146 namely zero for NaNs and the minimal or maximal value respectively 2147 for underflow or overflow. 2148 The *isExact output tells whether the result is exact, in the sense 2149 that converting it back to the original floating point type produces 2150 the original value. This is almost equivalent to result==opOK, 2151 except for negative zeroes. 2152 */ 2153 APFloat::opStatus 2154 APFloat::convertToInteger(integerPart *parts, unsigned int width, 2155 bool isSigned, 2156 roundingMode rounding_mode, bool *isExact) const 2157 { 2158 opStatus fs; 2159 2160 fs = convertToSignExtendedInteger(parts, width, isSigned, rounding_mode, 2161 isExact); 2162 2163 if (fs == opInvalidOp) { 2164 unsigned int bits, dstPartsCount; 2165 2166 dstPartsCount = partCountForBits(width); 2167 2168 if (category == fcNaN) 2169 bits = 0; 2170 else if (sign) 2171 bits = isSigned; 2172 else 2173 bits = width - isSigned; 2174 2175 APInt::tcSetLeastSignificantBits(parts, dstPartsCount, bits); 2176 if (sign && isSigned) 2177 APInt::tcShiftLeft(parts, dstPartsCount, width - 1); 2178 } 2179 2180 return fs; 2181 } 2182 2183 /* Same as convertToInteger(integerPart*, ...), except the result is returned in 2184 an APSInt, whose initial bit-width and signed-ness are used to determine the 2185 precision of the conversion. 2186 */ 2187 APFloat::opStatus 2188 APFloat::convertToInteger(APSInt &result, 2189 roundingMode rounding_mode, bool *isExact) const 2190 { 2191 unsigned bitWidth = result.getBitWidth(); 2192 SmallVector<uint64_t, 4> parts(result.getNumWords()); 2193 opStatus status = convertToInteger( 2194 parts.data(), bitWidth, result.isSigned(), rounding_mode, isExact); 2195 // Keeps the original signed-ness. 2196 result = APInt(bitWidth, parts); 2197 return status; 2198 } 2199 2200 /* Convert an unsigned integer SRC to a floating point number, 2201 rounding according to ROUNDING_MODE. The sign of the floating 2202 point number is not modified. */ 2203 APFloat::opStatus 2204 APFloat::convertFromUnsignedParts(const integerPart *src, 2205 unsigned int srcCount, 2206 roundingMode rounding_mode) 2207 { 2208 unsigned int omsb, precision, dstCount; 2209 integerPart *dst; 2210 lostFraction lost_fraction; 2211 2212 category = fcNormal; 2213 omsb = APInt::tcMSB(src, srcCount) + 1; 2214 dst = significandParts(); 2215 dstCount = partCount(); 2216 precision = semantics->precision; 2217 2218 /* We want the most significant PRECISION bits of SRC. There may not 2219 be that many; extract what we can. */ 2220 if (precision <= omsb) { 2221 exponent = omsb - 1; 2222 lost_fraction = lostFractionThroughTruncation(src, srcCount, 2223 omsb - precision); 2224 APInt::tcExtract(dst, dstCount, src, precision, omsb - precision); 2225 } else { 2226 exponent = precision - 1; 2227 lost_fraction = lfExactlyZero; 2228 APInt::tcExtract(dst, dstCount, src, omsb, 0); 2229 } 2230 2231 return normalize(rounding_mode, lost_fraction); 2232 } 2233 2234 APFloat::opStatus 2235 APFloat::convertFromAPInt(const APInt &Val, 2236 bool isSigned, 2237 roundingMode rounding_mode) 2238 { 2239 unsigned int partCount = Val.getNumWords(); 2240 APInt api = Val; 2241 2242 sign = false; 2243 if (isSigned && api.isNegative()) { 2244 sign = true; 2245 api = -api; 2246 } 2247 2248 return convertFromUnsignedParts(api.getRawData(), partCount, rounding_mode); 2249 } 2250 2251 /* Convert a two's complement integer SRC to a floating point number, 2252 rounding according to ROUNDING_MODE. ISSIGNED is true if the 2253 integer is signed, in which case it must be sign-extended. */ 2254 APFloat::opStatus 2255 APFloat::convertFromSignExtendedInteger(const integerPart *src, 2256 unsigned int srcCount, 2257 bool isSigned, 2258 roundingMode rounding_mode) 2259 { 2260 opStatus status; 2261 2262 if (isSigned && 2263 APInt::tcExtractBit(src, srcCount * integerPartWidth - 1)) { 2264 integerPart *copy; 2265 2266 /* If we're signed and negative negate a copy. */ 2267 sign = true; 2268 copy = new integerPart[srcCount]; 2269 APInt::tcAssign(copy, src, srcCount); 2270 APInt::tcNegate(copy, srcCount); 2271 status = convertFromUnsignedParts(copy, srcCount, rounding_mode); 2272 delete [] copy; 2273 } else { 2274 sign = false; 2275 status = convertFromUnsignedParts(src, srcCount, rounding_mode); 2276 } 2277 2278 return status; 2279 } 2280 2281 /* FIXME: should this just take a const APInt reference? */ 2282 APFloat::opStatus 2283 APFloat::convertFromZeroExtendedInteger(const integerPart *parts, 2284 unsigned int width, bool isSigned, 2285 roundingMode rounding_mode) 2286 { 2287 unsigned int partCount = partCountForBits(width); 2288 APInt api = APInt(width, makeArrayRef(parts, partCount)); 2289 2290 sign = false; 2291 if (isSigned && APInt::tcExtractBit(parts, width - 1)) { 2292 sign = true; 2293 api = -api; 2294 } 2295 2296 return convertFromUnsignedParts(api.getRawData(), partCount, rounding_mode); 2297 } 2298 2299 APFloat::opStatus 2300 APFloat::convertFromHexadecimalString(StringRef s, roundingMode rounding_mode) 2301 { 2302 lostFraction lost_fraction = lfExactlyZero; 2303 2304 zeroSignificand(); 2305 exponent = 0; 2306 category = fcNormal; 2307 2308 integerPart *significand = significandParts(); 2309 unsigned partsCount = partCount(); 2310 unsigned bitPos = partsCount * integerPartWidth; 2311 bool computedTrailingFraction = false; 2312 2313 // Skip leading zeroes and any (hexa)decimal point. 2314 StringRef::iterator begin = s.begin(); 2315 StringRef::iterator end = s.end(); 2316 StringRef::iterator dot; 2317 StringRef::iterator p = skipLeadingZeroesAndAnyDot(begin, end, &dot); 2318 StringRef::iterator firstSignificantDigit = p; 2319 2320 while (p != end) { 2321 integerPart hex_value; 2322 2323 if (*p == '.') { 2324 assert(dot == end && "String contains multiple dots"); 2325 dot = p++; 2326 continue; 2327 } 2328 2329 hex_value = hexDigitValue(*p); 2330 if (hex_value == -1U) 2331 break; 2332 2333 p++; 2334 2335 // Store the number while we have space. 2336 if (bitPos) { 2337 bitPos -= 4; 2338 hex_value <<= bitPos % integerPartWidth; 2339 significand[bitPos / integerPartWidth] |= hex_value; 2340 } else if (!computedTrailingFraction) { 2341 lost_fraction = trailingHexadecimalFraction(p, end, hex_value); 2342 computedTrailingFraction = true; 2343 } 2344 } 2345 2346 /* Hex floats require an exponent but not a hexadecimal point. */ 2347 assert(p != end && "Hex strings require an exponent"); 2348 assert((*p == 'p' || *p == 'P') && "Invalid character in significand"); 2349 assert(p != begin && "Significand has no digits"); 2350 assert((dot == end || p - begin != 1) && "Significand has no digits"); 2351 2352 /* Ignore the exponent if we are zero. */ 2353 if (p != firstSignificantDigit) { 2354 int expAdjustment; 2355 2356 /* Implicit hexadecimal point? */ 2357 if (dot == end) 2358 dot = p; 2359 2360 /* Calculate the exponent adjustment implicit in the number of 2361 significant digits. */ 2362 expAdjustment = static_cast<int>(dot - firstSignificantDigit); 2363 if (expAdjustment < 0) 2364 expAdjustment++; 2365 expAdjustment = expAdjustment * 4 - 1; 2366 2367 /* Adjust for writing the significand starting at the most 2368 significant nibble. */ 2369 expAdjustment += semantics->precision; 2370 expAdjustment -= partsCount * integerPartWidth; 2371 2372 /* Adjust for the given exponent. */ 2373 exponent = totalExponent(p + 1, end, expAdjustment); 2374 } 2375 2376 return normalize(rounding_mode, lost_fraction); 2377 } 2378 2379 APFloat::opStatus 2380 APFloat::roundSignificandWithExponent(const integerPart *decSigParts, 2381 unsigned sigPartCount, int exp, 2382 roundingMode rounding_mode) 2383 { 2384 unsigned int parts, pow5PartCount; 2385 fltSemantics calcSemantics = { 32767, -32767, 0 }; 2386 integerPart pow5Parts[maxPowerOfFiveParts]; 2387 bool isNearest; 2388 2389 isNearest = (rounding_mode == rmNearestTiesToEven || 2390 rounding_mode == rmNearestTiesToAway); 2391 2392 parts = partCountForBits(semantics->precision + 11); 2393 2394 /* Calculate pow(5, abs(exp)). */ 2395 pow5PartCount = powerOf5(pow5Parts, exp >= 0 ? exp: -exp); 2396 2397 for (;; parts *= 2) { 2398 opStatus sigStatus, powStatus; 2399 unsigned int excessPrecision, truncatedBits; 2400 2401 calcSemantics.precision = parts * integerPartWidth - 1; 2402 excessPrecision = calcSemantics.precision - semantics->precision; 2403 truncatedBits = excessPrecision; 2404 2405 APFloat decSig = APFloat::getZero(calcSemantics, sign); 2406 APFloat pow5(calcSemantics); 2407 2408 sigStatus = decSig.convertFromUnsignedParts(decSigParts, sigPartCount, 2409 rmNearestTiesToEven); 2410 powStatus = pow5.convertFromUnsignedParts(pow5Parts, pow5PartCount, 2411 rmNearestTiesToEven); 2412 /* Add exp, as 10^n = 5^n * 2^n. */ 2413 decSig.exponent += exp; 2414 2415 lostFraction calcLostFraction; 2416 integerPart HUerr, HUdistance; 2417 unsigned int powHUerr; 2418 2419 if (exp >= 0) { 2420 /* multiplySignificand leaves the precision-th bit set to 1. */ 2421 calcLostFraction = decSig.multiplySignificand(pow5, NULL); 2422 powHUerr = powStatus != opOK; 2423 } else { 2424 calcLostFraction = decSig.divideSignificand(pow5); 2425 /* Denormal numbers have less precision. */ 2426 if (decSig.exponent < semantics->minExponent) { 2427 excessPrecision += (semantics->minExponent - decSig.exponent); 2428 truncatedBits = excessPrecision; 2429 if (excessPrecision > calcSemantics.precision) 2430 excessPrecision = calcSemantics.precision; 2431 } 2432 /* Extra half-ulp lost in reciprocal of exponent. */ 2433 powHUerr = (powStatus == opOK && calcLostFraction == lfExactlyZero) ? 0:2; 2434 } 2435 2436 /* Both multiplySignificand and divideSignificand return the 2437 result with the integer bit set. */ 2438 assert(APInt::tcExtractBit 2439 (decSig.significandParts(), calcSemantics.precision - 1) == 1); 2440 2441 HUerr = HUerrBound(calcLostFraction != lfExactlyZero, sigStatus != opOK, 2442 powHUerr); 2443 HUdistance = 2 * ulpsFromBoundary(decSig.significandParts(), 2444 excessPrecision, isNearest); 2445 2446 /* Are we guaranteed to round correctly if we truncate? */ 2447 if (HUdistance >= HUerr) { 2448 APInt::tcExtract(significandParts(), partCount(), decSig.significandParts(), 2449 calcSemantics.precision - excessPrecision, 2450 excessPrecision); 2451 /* Take the exponent of decSig. If we tcExtract-ed less bits 2452 above we must adjust our exponent to compensate for the 2453 implicit right shift. */ 2454 exponent = (decSig.exponent + semantics->precision 2455 - (calcSemantics.precision - excessPrecision)); 2456 calcLostFraction = lostFractionThroughTruncation(decSig.significandParts(), 2457 decSig.partCount(), 2458 truncatedBits); 2459 return normalize(rounding_mode, calcLostFraction); 2460 } 2461 } 2462 } 2463 2464 APFloat::opStatus 2465 APFloat::convertFromDecimalString(StringRef str, roundingMode rounding_mode) 2466 { 2467 decimalInfo D; 2468 opStatus fs; 2469 2470 /* Scan the text. */ 2471 StringRef::iterator p = str.begin(); 2472 interpretDecimal(p, str.end(), &D); 2473 2474 /* Handle the quick cases. First the case of no significant digits, 2475 i.e. zero, and then exponents that are obviously too large or too 2476 small. Writing L for log 10 / log 2, a number d.ddddd*10^exp 2477 definitely overflows if 2478 2479 (exp - 1) * L >= maxExponent 2480 2481 and definitely underflows to zero where 2482 2483 (exp + 1) * L <= minExponent - precision 2484 2485 With integer arithmetic the tightest bounds for L are 2486 2487 93/28 < L < 196/59 [ numerator <= 256 ] 2488 42039/12655 < L < 28738/8651 [ numerator <= 65536 ] 2489 */ 2490 2491 // Test if we have a zero number allowing for strings with no null terminators 2492 // and zero decimals with non-zero exponents. 2493 // 2494 // We computed firstSigDigit by ignoring all zeros and dots. Thus if 2495 // D->firstSigDigit equals str.end(), every digit must be a zero and there can 2496 // be at most one dot. On the other hand, if we have a zero with a non-zero 2497 // exponent, then we know that D.firstSigDigit will be non-numeric. 2498 if (D.firstSigDigit == str.end() || decDigitValue(*D.firstSigDigit) >= 10U) { 2499 category = fcZero; 2500 fs = opOK; 2501 2502 /* Check whether the normalized exponent is high enough to overflow 2503 max during the log-rebasing in the max-exponent check below. */ 2504 } else if (D.normalizedExponent - 1 > INT_MAX / 42039) { 2505 fs = handleOverflow(rounding_mode); 2506 2507 /* If it wasn't, then it also wasn't high enough to overflow max 2508 during the log-rebasing in the min-exponent check. Check that it 2509 won't overflow min in either check, then perform the min-exponent 2510 check. */ 2511 } else if (D.normalizedExponent - 1 < INT_MIN / 42039 || 2512 (D.normalizedExponent + 1) * 28738 <= 2513 8651 * (semantics->minExponent - (int) semantics->precision)) { 2514 /* Underflow to zero and round. */ 2515 zeroSignificand(); 2516 fs = normalize(rounding_mode, lfLessThanHalf); 2517 2518 /* We can finally safely perform the max-exponent check. */ 2519 } else if ((D.normalizedExponent - 1) * 42039 2520 >= 12655 * semantics->maxExponent) { 2521 /* Overflow and round. */ 2522 fs = handleOverflow(rounding_mode); 2523 } else { 2524 integerPart *decSignificand; 2525 unsigned int partCount; 2526 2527 /* A tight upper bound on number of bits required to hold an 2528 N-digit decimal integer is N * 196 / 59. Allocate enough space 2529 to hold the full significand, and an extra part required by 2530 tcMultiplyPart. */ 2531 partCount = static_cast<unsigned int>(D.lastSigDigit - D.firstSigDigit) + 1; 2532 partCount = partCountForBits(1 + 196 * partCount / 59); 2533 decSignificand = new integerPart[partCount + 1]; 2534 partCount = 0; 2535 2536 /* Convert to binary efficiently - we do almost all multiplication 2537 in an integerPart. When this would overflow do we do a single 2538 bignum multiplication, and then revert again to multiplication 2539 in an integerPart. */ 2540 do { 2541 integerPart decValue, val, multiplier; 2542 2543 val = 0; 2544 multiplier = 1; 2545 2546 do { 2547 if (*p == '.') { 2548 p++; 2549 if (p == str.end()) { 2550 break; 2551 } 2552 } 2553 decValue = decDigitValue(*p++); 2554 assert(decValue < 10U && "Invalid character in significand"); 2555 multiplier *= 10; 2556 val = val * 10 + decValue; 2557 /* The maximum number that can be multiplied by ten with any 2558 digit added without overflowing an integerPart. */ 2559 } while (p <= D.lastSigDigit && multiplier <= (~ (integerPart) 0 - 9) / 10); 2560 2561 /* Multiply out the current part. */ 2562 APInt::tcMultiplyPart(decSignificand, decSignificand, multiplier, val, 2563 partCount, partCount + 1, false); 2564 2565 /* If we used another part (likely but not guaranteed), increase 2566 the count. */ 2567 if (decSignificand[partCount]) 2568 partCount++; 2569 } while (p <= D.lastSigDigit); 2570 2571 category = fcNormal; 2572 fs = roundSignificandWithExponent(decSignificand, partCount, 2573 D.exponent, rounding_mode); 2574 2575 delete [] decSignificand; 2576 } 2577 2578 return fs; 2579 } 2580 2581 bool 2582 APFloat::convertFromStringSpecials(StringRef str) { 2583 if (str.equals("inf") || str.equals("INFINITY")) { 2584 makeInf(false); 2585 return true; 2586 } 2587 2588 if (str.equals("-inf") || str.equals("-INFINITY")) { 2589 makeInf(true); 2590 return true; 2591 } 2592 2593 if (str.equals("nan") || str.equals("NaN")) { 2594 makeNaN(false, false); 2595 return true; 2596 } 2597 2598 if (str.equals("-nan") || str.equals("-NaN")) { 2599 makeNaN(false, true); 2600 return true; 2601 } 2602 2603 return false; 2604 } 2605 2606 APFloat::opStatus 2607 APFloat::convertFromString(StringRef str, roundingMode rounding_mode) 2608 { 2609 assert(!str.empty() && "Invalid string length"); 2610 2611 // Handle special cases. 2612 if (convertFromStringSpecials(str)) 2613 return opOK; 2614 2615 /* Handle a leading minus sign. */ 2616 StringRef::iterator p = str.begin(); 2617 size_t slen = str.size(); 2618 sign = *p == '-' ? 1 : 0; 2619 if (*p == '-' || *p == '+') { 2620 p++; 2621 slen--; 2622 assert(slen && "String has no digits"); 2623 } 2624 2625 if (slen >= 2 && p[0] == '0' && (p[1] == 'x' || p[1] == 'X')) { 2626 assert(slen - 2 && "Invalid string"); 2627 return convertFromHexadecimalString(StringRef(p + 2, slen - 2), 2628 rounding_mode); 2629 } 2630 2631 return convertFromDecimalString(StringRef(p, slen), rounding_mode); 2632 } 2633 2634 /* Write out a hexadecimal representation of the floating point value 2635 to DST, which must be of sufficient size, in the C99 form 2636 [-]0xh.hhhhp[+-]d. Return the number of characters written, 2637 excluding the terminating NUL. 2638 2639 If UPPERCASE, the output is in upper case, otherwise in lower case. 2640 2641 HEXDIGITS digits appear altogether, rounding the value if 2642 necessary. If HEXDIGITS is 0, the minimal precision to display the 2643 number precisely is used instead. If nothing would appear after 2644 the decimal point it is suppressed. 2645 2646 The decimal exponent is always printed and has at least one digit. 2647 Zero values display an exponent of zero. Infinities and NaNs 2648 appear as "infinity" or "nan" respectively. 2649 2650 The above rules are as specified by C99. There is ambiguity about 2651 what the leading hexadecimal digit should be. This implementation 2652 uses whatever is necessary so that the exponent is displayed as 2653 stored. This implies the exponent will fall within the IEEE format 2654 range, and the leading hexadecimal digit will be 0 (for denormals), 2655 1 (normal numbers) or 2 (normal numbers rounded-away-from-zero with 2656 any other digits zero). 2657 */ 2658 unsigned int 2659 APFloat::convertToHexString(char *dst, unsigned int hexDigits, 2660 bool upperCase, roundingMode rounding_mode) const 2661 { 2662 char *p; 2663 2664 p = dst; 2665 if (sign) 2666 *dst++ = '-'; 2667 2668 switch (category) { 2669 case fcInfinity: 2670 memcpy (dst, upperCase ? infinityU: infinityL, sizeof infinityU - 1); 2671 dst += sizeof infinityL - 1; 2672 break; 2673 2674 case fcNaN: 2675 memcpy (dst, upperCase ? NaNU: NaNL, sizeof NaNU - 1); 2676 dst += sizeof NaNU - 1; 2677 break; 2678 2679 case fcZero: 2680 *dst++ = '0'; 2681 *dst++ = upperCase ? 'X': 'x'; 2682 *dst++ = '0'; 2683 if (hexDigits > 1) { 2684 *dst++ = '.'; 2685 memset (dst, '0', hexDigits - 1); 2686 dst += hexDigits - 1; 2687 } 2688 *dst++ = upperCase ? 'P': 'p'; 2689 *dst++ = '0'; 2690 break; 2691 2692 case fcNormal: 2693 dst = convertNormalToHexString (dst, hexDigits, upperCase, rounding_mode); 2694 break; 2695 } 2696 2697 *dst = 0; 2698 2699 return static_cast<unsigned int>(dst - p); 2700 } 2701 2702 /* Does the hard work of outputting the correctly rounded hexadecimal 2703 form of a normal floating point number with the specified number of 2704 hexadecimal digits. If HEXDIGITS is zero the minimum number of 2705 digits necessary to print the value precisely is output. */ 2706 char * 2707 APFloat::convertNormalToHexString(char *dst, unsigned int hexDigits, 2708 bool upperCase, 2709 roundingMode rounding_mode) const 2710 { 2711 unsigned int count, valueBits, shift, partsCount, outputDigits; 2712 const char *hexDigitChars; 2713 const integerPart *significand; 2714 char *p; 2715 bool roundUp; 2716 2717 *dst++ = '0'; 2718 *dst++ = upperCase ? 'X': 'x'; 2719 2720 roundUp = false; 2721 hexDigitChars = upperCase ? hexDigitsUpper: hexDigitsLower; 2722 2723 significand = significandParts(); 2724 partsCount = partCount(); 2725 2726 /* +3 because the first digit only uses the single integer bit, so 2727 we have 3 virtual zero most-significant-bits. */ 2728 valueBits = semantics->precision + 3; 2729 shift = integerPartWidth - valueBits % integerPartWidth; 2730 2731 /* The natural number of digits required ignoring trailing 2732 insignificant zeroes. */ 2733 outputDigits = (valueBits - significandLSB () + 3) / 4; 2734 2735 /* hexDigits of zero means use the required number for the 2736 precision. Otherwise, see if we are truncating. If we are, 2737 find out if we need to round away from zero. */ 2738 if (hexDigits) { 2739 if (hexDigits < outputDigits) { 2740 /* We are dropping non-zero bits, so need to check how to round. 2741 "bits" is the number of dropped bits. */ 2742 unsigned int bits; 2743 lostFraction fraction; 2744 2745 bits = valueBits - hexDigits * 4; 2746 fraction = lostFractionThroughTruncation (significand, partsCount, bits); 2747 roundUp = roundAwayFromZero(rounding_mode, fraction, bits); 2748 } 2749 outputDigits = hexDigits; 2750 } 2751 2752 /* Write the digits consecutively, and start writing in the location 2753 of the hexadecimal point. We move the most significant digit 2754 left and add the hexadecimal point later. */ 2755 p = ++dst; 2756 2757 count = (valueBits + integerPartWidth - 1) / integerPartWidth; 2758 2759 while (outputDigits && count) { 2760 integerPart part; 2761 2762 /* Put the most significant integerPartWidth bits in "part". */ 2763 if (--count == partsCount) 2764 part = 0; /* An imaginary higher zero part. */ 2765 else 2766 part = significand[count] << shift; 2767 2768 if (count && shift) 2769 part |= significand[count - 1] >> (integerPartWidth - shift); 2770 2771 /* Convert as much of "part" to hexdigits as we can. */ 2772 unsigned int curDigits = integerPartWidth / 4; 2773 2774 if (curDigits > outputDigits) 2775 curDigits = outputDigits; 2776 dst += partAsHex (dst, part, curDigits, hexDigitChars); 2777 outputDigits -= curDigits; 2778 } 2779 2780 if (roundUp) { 2781 char *q = dst; 2782 2783 /* Note that hexDigitChars has a trailing '0'. */ 2784 do { 2785 q--; 2786 *q = hexDigitChars[hexDigitValue (*q) + 1]; 2787 } while (*q == '0'); 2788 assert(q >= p); 2789 } else { 2790 /* Add trailing zeroes. */ 2791 memset (dst, '0', outputDigits); 2792 dst += outputDigits; 2793 } 2794 2795 /* Move the most significant digit to before the point, and if there 2796 is something after the decimal point add it. This must come 2797 after rounding above. */ 2798 p[-1] = p[0]; 2799 if (dst -1 == p) 2800 dst--; 2801 else 2802 p[0] = '.'; 2803 2804 /* Finally output the exponent. */ 2805 *dst++ = upperCase ? 'P': 'p'; 2806 2807 return writeSignedDecimal (dst, exponent); 2808 } 2809 2810 hash_code llvm::hash_value(const APFloat &Arg) { 2811 if (!Arg.isFiniteNonZero()) 2812 return hash_combine((uint8_t)Arg.category, 2813 // NaN has no sign, fix it at zero. 2814 Arg.isNaN() ? (uint8_t)0 : (uint8_t)Arg.sign, 2815 Arg.semantics->precision); 2816 2817 // Normal floats need their exponent and significand hashed. 2818 return hash_combine((uint8_t)Arg.category, (uint8_t)Arg.sign, 2819 Arg.semantics->precision, Arg.exponent, 2820 hash_combine_range( 2821 Arg.significandParts(), 2822 Arg.significandParts() + Arg.partCount())); 2823 } 2824 2825 // Conversion from APFloat to/from host float/double. It may eventually be 2826 // possible to eliminate these and have everybody deal with APFloats, but that 2827 // will take a while. This approach will not easily extend to long double. 2828 // Current implementation requires integerPartWidth==64, which is correct at 2829 // the moment but could be made more general. 2830 2831 // Denormals have exponent minExponent in APFloat, but minExponent-1 in 2832 // the actual IEEE respresentations. We compensate for that here. 2833 2834 APInt 2835 APFloat::convertF80LongDoubleAPFloatToAPInt() const 2836 { 2837 assert(semantics == (const llvm::fltSemantics*)&x87DoubleExtended); 2838 assert(partCount()==2); 2839 2840 uint64_t myexponent, mysignificand; 2841 2842 if (isFiniteNonZero()) { 2843 myexponent = exponent+16383; //bias 2844 mysignificand = significandParts()[0]; 2845 if (myexponent==1 && !(mysignificand & 0x8000000000000000ULL)) 2846 myexponent = 0; // denormal 2847 } else if (category==fcZero) { 2848 myexponent = 0; 2849 mysignificand = 0; 2850 } else if (category==fcInfinity) { 2851 myexponent = 0x7fff; 2852 mysignificand = 0x8000000000000000ULL; 2853 } else { 2854 assert(category == fcNaN && "Unknown category"); 2855 myexponent = 0x7fff; 2856 mysignificand = significandParts()[0]; 2857 } 2858 2859 uint64_t words[2]; 2860 words[0] = mysignificand; 2861 words[1] = ((uint64_t)(sign & 1) << 15) | 2862 (myexponent & 0x7fffLL); 2863 return APInt(80, words); 2864 } 2865 2866 APInt 2867 APFloat::convertPPCDoubleDoubleAPFloatToAPInt() const 2868 { 2869 assert(semantics == (const llvm::fltSemantics*)&PPCDoubleDouble); 2870 assert(partCount()==2); 2871 2872 uint64_t words[2]; 2873 opStatus fs; 2874 bool losesInfo; 2875 2876 // Convert number to double. To avoid spurious underflows, we re- 2877 // normalize against the "double" minExponent first, and only *then* 2878 // truncate the mantissa. The result of that second conversion 2879 // may be inexact, but should never underflow. 2880 // Declare fltSemantics before APFloat that uses it (and 2881 // saves pointer to it) to ensure correct destruction order. 2882 fltSemantics extendedSemantics = *semantics; 2883 extendedSemantics.minExponent = IEEEdouble.minExponent; 2884 APFloat extended(*this); 2885 fs = extended.convert(extendedSemantics, rmNearestTiesToEven, &losesInfo); 2886 assert(fs == opOK && !losesInfo); 2887 (void)fs; 2888 2889 APFloat u(extended); 2890 fs = u.convert(IEEEdouble, rmNearestTiesToEven, &losesInfo); 2891 assert(fs == opOK || fs == opInexact); 2892 (void)fs; 2893 words[0] = *u.convertDoubleAPFloatToAPInt().getRawData(); 2894 2895 // If conversion was exact or resulted in a special case, we're done; 2896 // just set the second double to zero. Otherwise, re-convert back to 2897 // the extended format and compute the difference. This now should 2898 // convert exactly to double. 2899 if (u.isFiniteNonZero() && losesInfo) { 2900 fs = u.convert(extendedSemantics, rmNearestTiesToEven, &losesInfo); 2901 assert(fs == opOK && !losesInfo); 2902 (void)fs; 2903 2904 APFloat v(extended); 2905 v.subtract(u, rmNearestTiesToEven); 2906 fs = v.convert(IEEEdouble, rmNearestTiesToEven, &losesInfo); 2907 assert(fs == opOK && !losesInfo); 2908 (void)fs; 2909 words[1] = *v.convertDoubleAPFloatToAPInt().getRawData(); 2910 } else { 2911 words[1] = 0; 2912 } 2913 2914 return APInt(128, words); 2915 } 2916 2917 APInt 2918 APFloat::convertQuadrupleAPFloatToAPInt() const 2919 { 2920 assert(semantics == (const llvm::fltSemantics*)&IEEEquad); 2921 assert(partCount()==2); 2922 2923 uint64_t myexponent, mysignificand, mysignificand2; 2924 2925 if (isFiniteNonZero()) { 2926 myexponent = exponent+16383; //bias 2927 mysignificand = significandParts()[0]; 2928 mysignificand2 = significandParts()[1]; 2929 if (myexponent==1 && !(mysignificand2 & 0x1000000000000LL)) 2930 myexponent = 0; // denormal 2931 } else if (category==fcZero) { 2932 myexponent = 0; 2933 mysignificand = mysignificand2 = 0; 2934 } else if (category==fcInfinity) { 2935 myexponent = 0x7fff; 2936 mysignificand = mysignificand2 = 0; 2937 } else { 2938 assert(category == fcNaN && "Unknown category!"); 2939 myexponent = 0x7fff; 2940 mysignificand = significandParts()[0]; 2941 mysignificand2 = significandParts()[1]; 2942 } 2943 2944 uint64_t words[2]; 2945 words[0] = mysignificand; 2946 words[1] = ((uint64_t)(sign & 1) << 63) | 2947 ((myexponent & 0x7fff) << 48) | 2948 (mysignificand2 & 0xffffffffffffLL); 2949 2950 return APInt(128, words); 2951 } 2952 2953 APInt 2954 APFloat::convertDoubleAPFloatToAPInt() const 2955 { 2956 assert(semantics == (const llvm::fltSemantics*)&IEEEdouble); 2957 assert(partCount()==1); 2958 2959 uint64_t myexponent, mysignificand; 2960 2961 if (isFiniteNonZero()) { 2962 myexponent = exponent+1023; //bias 2963 mysignificand = *significandParts(); 2964 if (myexponent==1 && !(mysignificand & 0x10000000000000LL)) 2965 myexponent = 0; // denormal 2966 } else if (category==fcZero) { 2967 myexponent = 0; 2968 mysignificand = 0; 2969 } else if (category==fcInfinity) { 2970 myexponent = 0x7ff; 2971 mysignificand = 0; 2972 } else { 2973 assert(category == fcNaN && "Unknown category!"); 2974 myexponent = 0x7ff; 2975 mysignificand = *significandParts(); 2976 } 2977 2978 return APInt(64, ((((uint64_t)(sign & 1) << 63) | 2979 ((myexponent & 0x7ff) << 52) | 2980 (mysignificand & 0xfffffffffffffLL)))); 2981 } 2982 2983 APInt 2984 APFloat::convertFloatAPFloatToAPInt() const 2985 { 2986 assert(semantics == (const llvm::fltSemantics*)&IEEEsingle); 2987 assert(partCount()==1); 2988 2989 uint32_t myexponent, mysignificand; 2990 2991 if (isFiniteNonZero()) { 2992 myexponent = exponent+127; //bias 2993 mysignificand = (uint32_t)*significandParts(); 2994 if (myexponent == 1 && !(mysignificand & 0x800000)) 2995 myexponent = 0; // denormal 2996 } else if (category==fcZero) { 2997 myexponent = 0; 2998 mysignificand = 0; 2999 } else if (category==fcInfinity) { 3000 myexponent = 0xff; 3001 mysignificand = 0; 3002 } else { 3003 assert(category == fcNaN && "Unknown category!"); 3004 myexponent = 0xff; 3005 mysignificand = (uint32_t)*significandParts(); 3006 } 3007 3008 return APInt(32, (((sign&1) << 31) | ((myexponent&0xff) << 23) | 3009 (mysignificand & 0x7fffff))); 3010 } 3011 3012 APInt 3013 APFloat::convertHalfAPFloatToAPInt() const 3014 { 3015 assert(semantics == (const llvm::fltSemantics*)&IEEEhalf); 3016 assert(partCount()==1); 3017 3018 uint32_t myexponent, mysignificand; 3019 3020 if (isFiniteNonZero()) { 3021 myexponent = exponent+15; //bias 3022 mysignificand = (uint32_t)*significandParts(); 3023 if (myexponent == 1 && !(mysignificand & 0x400)) 3024 myexponent = 0; // denormal 3025 } else if (category==fcZero) { 3026 myexponent = 0; 3027 mysignificand = 0; 3028 } else if (category==fcInfinity) { 3029 myexponent = 0x1f; 3030 mysignificand = 0; 3031 } else { 3032 assert(category == fcNaN && "Unknown category!"); 3033 myexponent = 0x1f; 3034 mysignificand = (uint32_t)*significandParts(); 3035 } 3036 3037 return APInt(16, (((sign&1) << 15) | ((myexponent&0x1f) << 10) | 3038 (mysignificand & 0x3ff))); 3039 } 3040 3041 // This function creates an APInt that is just a bit map of the floating 3042 // point constant as it would appear in memory. It is not a conversion, 3043 // and treating the result as a normal integer is unlikely to be useful. 3044 3045 APInt 3046 APFloat::bitcastToAPInt() const 3047 { 3048 if (semantics == (const llvm::fltSemantics*)&IEEEhalf) 3049 return convertHalfAPFloatToAPInt(); 3050 3051 if (semantics == (const llvm::fltSemantics*)&IEEEsingle) 3052 return convertFloatAPFloatToAPInt(); 3053 3054 if (semantics == (const llvm::fltSemantics*)&IEEEdouble) 3055 return convertDoubleAPFloatToAPInt(); 3056 3057 if (semantics == (const llvm::fltSemantics*)&IEEEquad) 3058 return convertQuadrupleAPFloatToAPInt(); 3059 3060 if (semantics == (const llvm::fltSemantics*)&PPCDoubleDouble) 3061 return convertPPCDoubleDoubleAPFloatToAPInt(); 3062 3063 assert(semantics == (const llvm::fltSemantics*)&x87DoubleExtended && 3064 "unknown format!"); 3065 return convertF80LongDoubleAPFloatToAPInt(); 3066 } 3067 3068 float 3069 APFloat::convertToFloat() const 3070 { 3071 assert(semantics == (const llvm::fltSemantics*)&IEEEsingle && 3072 "Float semantics are not IEEEsingle"); 3073 APInt api = bitcastToAPInt(); 3074 return api.bitsToFloat(); 3075 } 3076 3077 double 3078 APFloat::convertToDouble() const 3079 { 3080 assert(semantics == (const llvm::fltSemantics*)&IEEEdouble && 3081 "Float semantics are not IEEEdouble"); 3082 APInt api = bitcastToAPInt(); 3083 return api.bitsToDouble(); 3084 } 3085 3086 /// Integer bit is explicit in this format. Intel hardware (387 and later) 3087 /// does not support these bit patterns: 3088 /// exponent = all 1's, integer bit 0, significand 0 ("pseudoinfinity") 3089 /// exponent = all 1's, integer bit 0, significand nonzero ("pseudoNaN") 3090 /// exponent = 0, integer bit 1 ("pseudodenormal") 3091 /// exponent!=0 nor all 1's, integer bit 0 ("unnormal") 3092 /// At the moment, the first two are treated as NaNs, the second two as Normal. 3093 void 3094 APFloat::initFromF80LongDoubleAPInt(const APInt &api) 3095 { 3096 assert(api.getBitWidth()==80); 3097 uint64_t i1 = api.getRawData()[0]; 3098 uint64_t i2 = api.getRawData()[1]; 3099 uint64_t myexponent = (i2 & 0x7fff); 3100 uint64_t mysignificand = i1; 3101 3102 initialize(&APFloat::x87DoubleExtended); 3103 assert(partCount()==2); 3104 3105 sign = static_cast<unsigned int>(i2>>15); 3106 if (myexponent==0 && mysignificand==0) { 3107 // exponent, significand meaningless 3108 category = fcZero; 3109 } else if (myexponent==0x7fff && mysignificand==0x8000000000000000ULL) { 3110 // exponent, significand meaningless 3111 category = fcInfinity; 3112 } else if (myexponent==0x7fff && mysignificand!=0x8000000000000000ULL) { 3113 // exponent meaningless 3114 category = fcNaN; 3115 significandParts()[0] = mysignificand; 3116 significandParts()[1] = 0; 3117 } else { 3118 category = fcNormal; 3119 exponent = myexponent - 16383; 3120 significandParts()[0] = mysignificand; 3121 significandParts()[1] = 0; 3122 if (myexponent==0) // denormal 3123 exponent = -16382; 3124 } 3125 } 3126 3127 void 3128 APFloat::initFromPPCDoubleDoubleAPInt(const APInt &api) 3129 { 3130 assert(api.getBitWidth()==128); 3131 uint64_t i1 = api.getRawData()[0]; 3132 uint64_t i2 = api.getRawData()[1]; 3133 opStatus fs; 3134 bool losesInfo; 3135 3136 // Get the first double and convert to our format. 3137 initFromDoubleAPInt(APInt(64, i1)); 3138 fs = convert(PPCDoubleDouble, rmNearestTiesToEven, &losesInfo); 3139 assert(fs == opOK && !losesInfo); 3140 (void)fs; 3141 3142 // Unless we have a special case, add in second double. 3143 if (isFiniteNonZero()) { 3144 APFloat v(IEEEdouble, APInt(64, i2)); 3145 fs = v.convert(PPCDoubleDouble, rmNearestTiesToEven, &losesInfo); 3146 assert(fs == opOK && !losesInfo); 3147 (void)fs; 3148 3149 add(v, rmNearestTiesToEven); 3150 } 3151 } 3152 3153 void 3154 APFloat::initFromQuadrupleAPInt(const APInt &api) 3155 { 3156 assert(api.getBitWidth()==128); 3157 uint64_t i1 = api.getRawData()[0]; 3158 uint64_t i2 = api.getRawData()[1]; 3159 uint64_t myexponent = (i2 >> 48) & 0x7fff; 3160 uint64_t mysignificand = i1; 3161 uint64_t mysignificand2 = i2 & 0xffffffffffffLL; 3162 3163 initialize(&APFloat::IEEEquad); 3164 assert(partCount()==2); 3165 3166 sign = static_cast<unsigned int>(i2>>63); 3167 if (myexponent==0 && 3168 (mysignificand==0 && mysignificand2==0)) { 3169 // exponent, significand meaningless 3170 category = fcZero; 3171 } else if (myexponent==0x7fff && 3172 (mysignificand==0 && mysignificand2==0)) { 3173 // exponent, significand meaningless 3174 category = fcInfinity; 3175 } else if (myexponent==0x7fff && 3176 (mysignificand!=0 || mysignificand2 !=0)) { 3177 // exponent meaningless 3178 category = fcNaN; 3179 significandParts()[0] = mysignificand; 3180 significandParts()[1] = mysignificand2; 3181 } else { 3182 category = fcNormal; 3183 exponent = myexponent - 16383; 3184 significandParts()[0] = mysignificand; 3185 significandParts()[1] = mysignificand2; 3186 if (myexponent==0) // denormal 3187 exponent = -16382; 3188 else 3189 significandParts()[1] |= 0x1000000000000LL; // integer bit 3190 } 3191 } 3192 3193 void 3194 APFloat::initFromDoubleAPInt(const APInt &api) 3195 { 3196 assert(api.getBitWidth()==64); 3197 uint64_t i = *api.getRawData(); 3198 uint64_t myexponent = (i >> 52) & 0x7ff; 3199 uint64_t mysignificand = i & 0xfffffffffffffLL; 3200 3201 initialize(&APFloat::IEEEdouble); 3202 assert(partCount()==1); 3203 3204 sign = static_cast<unsigned int>(i>>63); 3205 if (myexponent==0 && mysignificand==0) { 3206 // exponent, significand meaningless 3207 category = fcZero; 3208 } else if (myexponent==0x7ff && mysignificand==0) { 3209 // exponent, significand meaningless 3210 category = fcInfinity; 3211 } else if (myexponent==0x7ff && mysignificand!=0) { 3212 // exponent meaningless 3213 category = fcNaN; 3214 *significandParts() = mysignificand; 3215 } else { 3216 category = fcNormal; 3217 exponent = myexponent - 1023; 3218 *significandParts() = mysignificand; 3219 if (myexponent==0) // denormal 3220 exponent = -1022; 3221 else 3222 *significandParts() |= 0x10000000000000LL; // integer bit 3223 } 3224 } 3225 3226 void 3227 APFloat::initFromFloatAPInt(const APInt & api) 3228 { 3229 assert(api.getBitWidth()==32); 3230 uint32_t i = (uint32_t)*api.getRawData(); 3231 uint32_t myexponent = (i >> 23) & 0xff; 3232 uint32_t mysignificand = i & 0x7fffff; 3233 3234 initialize(&APFloat::IEEEsingle); 3235 assert(partCount()==1); 3236 3237 sign = i >> 31; 3238 if (myexponent==0 && mysignificand==0) { 3239 // exponent, significand meaningless 3240 category = fcZero; 3241 } else if (myexponent==0xff && mysignificand==0) { 3242 // exponent, significand meaningless 3243 category = fcInfinity; 3244 } else if (myexponent==0xff && mysignificand!=0) { 3245 // sign, exponent, significand meaningless 3246 category = fcNaN; 3247 *significandParts() = mysignificand; 3248 } else { 3249 category = fcNormal; 3250 exponent = myexponent - 127; //bias 3251 *significandParts() = mysignificand; 3252 if (myexponent==0) // denormal 3253 exponent = -126; 3254 else 3255 *significandParts() |= 0x800000; // integer bit 3256 } 3257 } 3258 3259 void 3260 APFloat::initFromHalfAPInt(const APInt & api) 3261 { 3262 assert(api.getBitWidth()==16); 3263 uint32_t i = (uint32_t)*api.getRawData(); 3264 uint32_t myexponent = (i >> 10) & 0x1f; 3265 uint32_t mysignificand = i & 0x3ff; 3266 3267 initialize(&APFloat::IEEEhalf); 3268 assert(partCount()==1); 3269 3270 sign = i >> 15; 3271 if (myexponent==0 && mysignificand==0) { 3272 // exponent, significand meaningless 3273 category = fcZero; 3274 } else if (myexponent==0x1f && mysignificand==0) { 3275 // exponent, significand meaningless 3276 category = fcInfinity; 3277 } else if (myexponent==0x1f && mysignificand!=0) { 3278 // sign, exponent, significand meaningless 3279 category = fcNaN; 3280 *significandParts() = mysignificand; 3281 } else { 3282 category = fcNormal; 3283 exponent = myexponent - 15; //bias 3284 *significandParts() = mysignificand; 3285 if (myexponent==0) // denormal 3286 exponent = -14; 3287 else 3288 *significandParts() |= 0x400; // integer bit 3289 } 3290 } 3291 3292 /// Treat api as containing the bits of a floating point number. Currently 3293 /// we infer the floating point type from the size of the APInt. The 3294 /// isIEEE argument distinguishes between PPC128 and IEEE128 (not meaningful 3295 /// when the size is anything else). 3296 void 3297 APFloat::initFromAPInt(const fltSemantics* Sem, const APInt& api) 3298 { 3299 if (Sem == &IEEEhalf) 3300 return initFromHalfAPInt(api); 3301 if (Sem == &IEEEsingle) 3302 return initFromFloatAPInt(api); 3303 if (Sem == &IEEEdouble) 3304 return initFromDoubleAPInt(api); 3305 if (Sem == &x87DoubleExtended) 3306 return initFromF80LongDoubleAPInt(api); 3307 if (Sem == &IEEEquad) 3308 return initFromQuadrupleAPInt(api); 3309 if (Sem == &PPCDoubleDouble) 3310 return initFromPPCDoubleDoubleAPInt(api); 3311 3312 llvm_unreachable(0); 3313 } 3314 3315 APFloat 3316 APFloat::getAllOnesValue(unsigned BitWidth, bool isIEEE) 3317 { 3318 switch (BitWidth) { 3319 case 16: 3320 return APFloat(IEEEhalf, APInt::getAllOnesValue(BitWidth)); 3321 case 32: 3322 return APFloat(IEEEsingle, APInt::getAllOnesValue(BitWidth)); 3323 case 64: 3324 return APFloat(IEEEdouble, APInt::getAllOnesValue(BitWidth)); 3325 case 80: 3326 return APFloat(x87DoubleExtended, APInt::getAllOnesValue(BitWidth)); 3327 case 128: 3328 if (isIEEE) 3329 return APFloat(IEEEquad, APInt::getAllOnesValue(BitWidth)); 3330 return APFloat(PPCDoubleDouble, APInt::getAllOnesValue(BitWidth)); 3331 default: 3332 llvm_unreachable("Unknown floating bit width"); 3333 } 3334 } 3335 3336 /// Make this number the largest magnitude normal number in the given 3337 /// semantics. 3338 void APFloat::makeLargest(bool Negative) { 3339 // We want (in interchange format): 3340 // sign = {Negative} 3341 // exponent = 1..10 3342 // significand = 1..1 3343 category = fcNormal; 3344 sign = Negative; 3345 exponent = semantics->maxExponent; 3346 3347 // Use memset to set all but the highest integerPart to all ones. 3348 integerPart *significand = significandParts(); 3349 unsigned PartCount = partCount(); 3350 memset(significand, 0xFF, sizeof(integerPart)*(PartCount - 1)); 3351 3352 // Set the high integerPart especially setting all unused top bits for 3353 // internal consistency. 3354 const unsigned NumUnusedHighBits = 3355 PartCount*integerPartWidth - semantics->precision; 3356 significand[PartCount - 1] = ~integerPart(0) >> NumUnusedHighBits; 3357 } 3358 3359 /// Make this number the smallest magnitude denormal number in the given 3360 /// semantics. 3361 void APFloat::makeSmallest(bool Negative) { 3362 // We want (in interchange format): 3363 // sign = {Negative} 3364 // exponent = 0..0 3365 // significand = 0..01 3366 category = fcNormal; 3367 sign = Negative; 3368 exponent = semantics->minExponent; 3369 APInt::tcSet(significandParts(), 1, partCount()); 3370 } 3371 3372 3373 APFloat APFloat::getLargest(const fltSemantics &Sem, bool Negative) { 3374 // We want (in interchange format): 3375 // sign = {Negative} 3376 // exponent = 1..10 3377 // significand = 1..1 3378 APFloat Val(Sem, uninitialized); 3379 Val.makeLargest(Negative); 3380 return Val; 3381 } 3382 3383 APFloat APFloat::getSmallest(const fltSemantics &Sem, bool Negative) { 3384 // We want (in interchange format): 3385 // sign = {Negative} 3386 // exponent = 0..0 3387 // significand = 0..01 3388 APFloat Val(Sem, uninitialized); 3389 Val.makeSmallest(Negative); 3390 return Val; 3391 } 3392 3393 APFloat APFloat::getSmallestNormalized(const fltSemantics &Sem, bool Negative) { 3394 APFloat Val(Sem, uninitialized); 3395 3396 // We want (in interchange format): 3397 // sign = {Negative} 3398 // exponent = 0..0 3399 // significand = 10..0 3400 3401 Val.zeroSignificand(); 3402 Val.sign = Negative; 3403 Val.exponent = Sem.minExponent; 3404 Val.significandParts()[partCountForBits(Sem.precision)-1] |= 3405 (((integerPart) 1) << ((Sem.precision - 1) % integerPartWidth)); 3406 3407 return Val; 3408 } 3409 3410 APFloat::APFloat(const fltSemantics &Sem, const APInt &API) { 3411 initFromAPInt(&Sem, API); 3412 } 3413 3414 APFloat::APFloat(float f) { 3415 initFromAPInt(&IEEEsingle, APInt::floatToBits(f)); 3416 } 3417 3418 APFloat::APFloat(double d) { 3419 initFromAPInt(&IEEEdouble, APInt::doubleToBits(d)); 3420 } 3421 3422 namespace { 3423 void append(SmallVectorImpl<char> &Buffer, StringRef Str) { 3424 Buffer.append(Str.begin(), Str.end()); 3425 } 3426 3427 /// Removes data from the given significand until it is no more 3428 /// precise than is required for the desired precision. 3429 void AdjustToPrecision(APInt &significand, 3430 int &exp, unsigned FormatPrecision) { 3431 unsigned bits = significand.getActiveBits(); 3432 3433 // 196/59 is a very slight overestimate of lg_2(10). 3434 unsigned bitsRequired = (FormatPrecision * 196 + 58) / 59; 3435 3436 if (bits <= bitsRequired) return; 3437 3438 unsigned tensRemovable = (bits - bitsRequired) * 59 / 196; 3439 if (!tensRemovable) return; 3440 3441 exp += tensRemovable; 3442 3443 APInt divisor(significand.getBitWidth(), 1); 3444 APInt powten(significand.getBitWidth(), 10); 3445 while (true) { 3446 if (tensRemovable & 1) 3447 divisor *= powten; 3448 tensRemovable >>= 1; 3449 if (!tensRemovable) break; 3450 powten *= powten; 3451 } 3452 3453 significand = significand.udiv(divisor); 3454 3455 // Truncate the significand down to its active bit count. 3456 significand = significand.trunc(significand.getActiveBits()); 3457 } 3458 3459 3460 void AdjustToPrecision(SmallVectorImpl<char> &buffer, 3461 int &exp, unsigned FormatPrecision) { 3462 unsigned N = buffer.size(); 3463 if (N <= FormatPrecision) return; 3464 3465 // The most significant figures are the last ones in the buffer. 3466 unsigned FirstSignificant = N - FormatPrecision; 3467 3468 // Round. 3469 // FIXME: this probably shouldn't use 'round half up'. 3470 3471 // Rounding down is just a truncation, except we also want to drop 3472 // trailing zeros from the new result. 3473 if (buffer[FirstSignificant - 1] < '5') { 3474 while (FirstSignificant < N && buffer[FirstSignificant] == '0') 3475 FirstSignificant++; 3476 3477 exp += FirstSignificant; 3478 buffer.erase(&buffer[0], &buffer[FirstSignificant]); 3479 return; 3480 } 3481 3482 // Rounding up requires a decimal add-with-carry. If we continue 3483 // the carry, the newly-introduced zeros will just be truncated. 3484 for (unsigned I = FirstSignificant; I != N; ++I) { 3485 if (buffer[I] == '9') { 3486 FirstSignificant++; 3487 } else { 3488 buffer[I]++; 3489 break; 3490 } 3491 } 3492 3493 // If we carried through, we have exactly one digit of precision. 3494 if (FirstSignificant == N) { 3495 exp += FirstSignificant; 3496 buffer.clear(); 3497 buffer.push_back('1'); 3498 return; 3499 } 3500 3501 exp += FirstSignificant; 3502 buffer.erase(&buffer[0], &buffer[FirstSignificant]); 3503 } 3504 } 3505 3506 void APFloat::toString(SmallVectorImpl<char> &Str, 3507 unsigned FormatPrecision, 3508 unsigned FormatMaxPadding) const { 3509 switch (category) { 3510 case fcInfinity: 3511 if (isNegative()) 3512 return append(Str, "-Inf"); 3513 else 3514 return append(Str, "+Inf"); 3515 3516 case fcNaN: return append(Str, "NaN"); 3517 3518 case fcZero: 3519 if (isNegative()) 3520 Str.push_back('-'); 3521 3522 if (!FormatMaxPadding) 3523 append(Str, "0.0E+0"); 3524 else 3525 Str.push_back('0'); 3526 return; 3527 3528 case fcNormal: 3529 break; 3530 } 3531 3532 if (isNegative()) 3533 Str.push_back('-'); 3534 3535 // Decompose the number into an APInt and an exponent. 3536 int exp = exponent - ((int) semantics->precision - 1); 3537 APInt significand(semantics->precision, 3538 makeArrayRef(significandParts(), 3539 partCountForBits(semantics->precision))); 3540 3541 // Set FormatPrecision if zero. We want to do this before we 3542 // truncate trailing zeros, as those are part of the precision. 3543 if (!FormatPrecision) { 3544 // It's an interesting question whether to use the nominal 3545 // precision or the active precision here for denormals. 3546 3547 // FormatPrecision = ceil(significandBits / lg_2(10)) 3548 FormatPrecision = (semantics->precision * 59 + 195) / 196; 3549 } 3550 3551 // Ignore trailing binary zeros. 3552 int trailingZeros = significand.countTrailingZeros(); 3553 exp += trailingZeros; 3554 significand = significand.lshr(trailingZeros); 3555 3556 // Change the exponent from 2^e to 10^e. 3557 if (exp == 0) { 3558 // Nothing to do. 3559 } else if (exp > 0) { 3560 // Just shift left. 3561 significand = significand.zext(semantics->precision + exp); 3562 significand <<= exp; 3563 exp = 0; 3564 } else { /* exp < 0 */ 3565 int texp = -exp; 3566 3567 // We transform this using the identity: 3568 // (N)(2^-e) == (N)(5^e)(10^-e) 3569 // This means we have to multiply N (the significand) by 5^e. 3570 // To avoid overflow, we have to operate on numbers large 3571 // enough to store N * 5^e: 3572 // log2(N * 5^e) == log2(N) + e * log2(5) 3573 // <= semantics->precision + e * 137 / 59 3574 // (log_2(5) ~ 2.321928 < 2.322034 ~ 137/59) 3575 3576 unsigned precision = semantics->precision + (137 * texp + 136) / 59; 3577 3578 // Multiply significand by 5^e. 3579 // N * 5^0101 == N * 5^(1*1) * 5^(0*2) * 5^(1*4) * 5^(0*8) 3580 significand = significand.zext(precision); 3581 APInt five_to_the_i(precision, 5); 3582 while (true) { 3583 if (texp & 1) significand *= five_to_the_i; 3584 3585 texp >>= 1; 3586 if (!texp) break; 3587 five_to_the_i *= five_to_the_i; 3588 } 3589 } 3590 3591 AdjustToPrecision(significand, exp, FormatPrecision); 3592 3593 SmallVector<char, 256> buffer; 3594 3595 // Fill the buffer. 3596 unsigned precision = significand.getBitWidth(); 3597 APInt ten(precision, 10); 3598 APInt digit(precision, 0); 3599 3600 bool inTrail = true; 3601 while (significand != 0) { 3602 // digit <- significand % 10 3603 // significand <- significand / 10 3604 APInt::udivrem(significand, ten, significand, digit); 3605 3606 unsigned d = digit.getZExtValue(); 3607 3608 // Drop trailing zeros. 3609 if (inTrail && !d) exp++; 3610 else { 3611 buffer.push_back((char) ('0' + d)); 3612 inTrail = false; 3613 } 3614 } 3615 3616 assert(!buffer.empty() && "no characters in buffer!"); 3617 3618 // Drop down to FormatPrecision. 3619 // TODO: don't do more precise calculations above than are required. 3620 AdjustToPrecision(buffer, exp, FormatPrecision); 3621 3622 unsigned NDigits = buffer.size(); 3623 3624 // Check whether we should use scientific notation. 3625 bool FormatScientific; 3626 if (!FormatMaxPadding) 3627 FormatScientific = true; 3628 else { 3629 if (exp >= 0) { 3630 // 765e3 --> 765000 3631 // ^^^ 3632 // But we shouldn't make the number look more precise than it is. 3633 FormatScientific = ((unsigned) exp > FormatMaxPadding || 3634 NDigits + (unsigned) exp > FormatPrecision); 3635 } else { 3636 // Power of the most significant digit. 3637 int MSD = exp + (int) (NDigits - 1); 3638 if (MSD >= 0) { 3639 // 765e-2 == 7.65 3640 FormatScientific = false; 3641 } else { 3642 // 765e-5 == 0.00765 3643 // ^ ^^ 3644 FormatScientific = ((unsigned) -MSD) > FormatMaxPadding; 3645 } 3646 } 3647 } 3648 3649 // Scientific formatting is pretty straightforward. 3650 if (FormatScientific) { 3651 exp += (NDigits - 1); 3652 3653 Str.push_back(buffer[NDigits-1]); 3654 Str.push_back('.'); 3655 if (NDigits == 1) 3656 Str.push_back('0'); 3657 else 3658 for (unsigned I = 1; I != NDigits; ++I) 3659 Str.push_back(buffer[NDigits-1-I]); 3660 Str.push_back('E'); 3661 3662 Str.push_back(exp >= 0 ? '+' : '-'); 3663 if (exp < 0) exp = -exp; 3664 SmallVector<char, 6> expbuf; 3665 do { 3666 expbuf.push_back((char) ('0' + (exp % 10))); 3667 exp /= 10; 3668 } while (exp); 3669 for (unsigned I = 0, E = expbuf.size(); I != E; ++I) 3670 Str.push_back(expbuf[E-1-I]); 3671 return; 3672 } 3673 3674 // Non-scientific, positive exponents. 3675 if (exp >= 0) { 3676 for (unsigned I = 0; I != NDigits; ++I) 3677 Str.push_back(buffer[NDigits-1-I]); 3678 for (unsigned I = 0; I != (unsigned) exp; ++I) 3679 Str.push_back('0'); 3680 return; 3681 } 3682 3683 // Non-scientific, negative exponents. 3684 3685 // The number of digits to the left of the decimal point. 3686 int NWholeDigits = exp + (int) NDigits; 3687 3688 unsigned I = 0; 3689 if (NWholeDigits > 0) { 3690 for (; I != (unsigned) NWholeDigits; ++I) 3691 Str.push_back(buffer[NDigits-I-1]); 3692 Str.push_back('.'); 3693 } else { 3694 unsigned NZeros = 1 + (unsigned) -NWholeDigits; 3695 3696 Str.push_back('0'); 3697 Str.push_back('.'); 3698 for (unsigned Z = 1; Z != NZeros; ++Z) 3699 Str.push_back('0'); 3700 } 3701 3702 for (; I != NDigits; ++I) 3703 Str.push_back(buffer[NDigits-I-1]); 3704 } 3705 3706 bool APFloat::getExactInverse(APFloat *inv) const { 3707 // Special floats and denormals have no exact inverse. 3708 if (!isFiniteNonZero()) 3709 return false; 3710 3711 // Check that the number is a power of two by making sure that only the 3712 // integer bit is set in the significand. 3713 if (significandLSB() != semantics->precision - 1) 3714 return false; 3715 3716 // Get the inverse. 3717 APFloat reciprocal(*semantics, 1ULL); 3718 if (reciprocal.divide(*this, rmNearestTiesToEven) != opOK) 3719 return false; 3720 3721 // Avoid multiplication with a denormal, it is not safe on all platforms and 3722 // may be slower than a normal division. 3723 if (reciprocal.isDenormal()) 3724 return false; 3725 3726 assert(reciprocal.isFiniteNonZero() && 3727 reciprocal.significandLSB() == reciprocal.semantics->precision - 1); 3728 3729 if (inv) 3730 *inv = reciprocal; 3731 3732 return true; 3733 } 3734 3735 bool APFloat::isSignaling() const { 3736 if (!isNaN()) 3737 return false; 3738 3739 // IEEE-754R 2008 6.2.1: A signaling NaN bit string should be encoded with the 3740 // first bit of the trailing significand being 0. 3741 return !APInt::tcExtractBit(significandParts(), semantics->precision - 2); 3742 } 3743 3744 /// IEEE-754R 2008 5.3.1: nextUp/nextDown. 3745 /// 3746 /// *NOTE* since nextDown(x) = -nextUp(-x), we only implement nextUp with 3747 /// appropriate sign switching before/after the computation. 3748 APFloat::opStatus APFloat::next(bool nextDown) { 3749 // If we are performing nextDown, swap sign so we have -x. 3750 if (nextDown) 3751 changeSign(); 3752 3753 // Compute nextUp(x) 3754 opStatus result = opOK; 3755 3756 // Handle each float category separately. 3757 switch (category) { 3758 case fcInfinity: 3759 // nextUp(+inf) = +inf 3760 if (!isNegative()) 3761 break; 3762 // nextUp(-inf) = -getLargest() 3763 makeLargest(true); 3764 break; 3765 case fcNaN: 3766 // IEEE-754R 2008 6.2 Par 2: nextUp(sNaN) = qNaN. Set Invalid flag. 3767 // IEEE-754R 2008 6.2: nextUp(qNaN) = qNaN. Must be identity so we do not 3768 // change the payload. 3769 if (isSignaling()) { 3770 result = opInvalidOp; 3771 // For consistency, propogate the sign of the sNaN to the qNaN. 3772 makeNaN(false, isNegative(), 0); 3773 } 3774 break; 3775 case fcZero: 3776 // nextUp(pm 0) = +getSmallest() 3777 makeSmallest(false); 3778 break; 3779 case fcNormal: 3780 // nextUp(-getSmallest()) = -0 3781 if (isSmallest() && isNegative()) { 3782 APInt::tcSet(significandParts(), 0, partCount()); 3783 category = fcZero; 3784 exponent = 0; 3785 break; 3786 } 3787 3788 // nextUp(getLargest()) == INFINITY 3789 if (isLargest() && !isNegative()) { 3790 APInt::tcSet(significandParts(), 0, partCount()); 3791 category = fcInfinity; 3792 exponent = semantics->maxExponent + 1; 3793 break; 3794 } 3795 3796 // nextUp(normal) == normal + inc. 3797 if (isNegative()) { 3798 // If we are negative, we need to decrement the significand. 3799 3800 // We only cross a binade boundary that requires adjusting the exponent 3801 // if: 3802 // 1. exponent != semantics->minExponent. This implies we are not in the 3803 // smallest binade or are dealing with denormals. 3804 // 2. Our significand excluding the integral bit is all zeros. 3805 bool WillCrossBinadeBoundary = 3806 exponent != semantics->minExponent && isSignificandAllZeros(); 3807 3808 // Decrement the significand. 3809 // 3810 // We always do this since: 3811 // 1. If we are dealing with a non binade decrement, by definition we 3812 // just decrement the significand. 3813 // 2. If we are dealing with a normal -> normal binade decrement, since 3814 // we have an explicit integral bit the fact that all bits but the 3815 // integral bit are zero implies that subtracting one will yield a 3816 // significand with 0 integral bit and 1 in all other spots. Thus we 3817 // must just adjust the exponent and set the integral bit to 1. 3818 // 3. If we are dealing with a normal -> denormal binade decrement, 3819 // since we set the integral bit to 0 when we represent denormals, we 3820 // just decrement the significand. 3821 integerPart *Parts = significandParts(); 3822 APInt::tcDecrement(Parts, partCount()); 3823 3824 if (WillCrossBinadeBoundary) { 3825 // Our result is a normal number. Do the following: 3826 // 1. Set the integral bit to 1. 3827 // 2. Decrement the exponent. 3828 APInt::tcSetBit(Parts, semantics->precision - 1); 3829 exponent--; 3830 } 3831 } else { 3832 // If we are positive, we need to increment the significand. 3833 3834 // We only cross a binade boundary that requires adjusting the exponent if 3835 // the input is not a denormal and all of said input's significand bits 3836 // are set. If all of said conditions are true: clear the significand, set 3837 // the integral bit to 1, and increment the exponent. If we have a 3838 // denormal always increment since moving denormals and the numbers in the 3839 // smallest normal binade have the same exponent in our representation. 3840 bool WillCrossBinadeBoundary = !isDenormal() && isSignificandAllOnes(); 3841 3842 if (WillCrossBinadeBoundary) { 3843 integerPart *Parts = significandParts(); 3844 APInt::tcSet(Parts, 0, partCount()); 3845 APInt::tcSetBit(Parts, semantics->precision - 1); 3846 assert(exponent != semantics->maxExponent && 3847 "We can not increment an exponent beyond the maxExponent allowed" 3848 " by the given floating point semantics."); 3849 exponent++; 3850 } else { 3851 incrementSignificand(); 3852 } 3853 } 3854 break; 3855 } 3856 3857 // If we are performing nextDown, swap sign so we have -nextUp(-x) 3858 if (nextDown) 3859 changeSign(); 3860 3861 return result; 3862 } 3863 3864 void 3865 APFloat::makeInf(bool Negative) { 3866 category = fcInfinity; 3867 sign = Negative; 3868 exponent = semantics->maxExponent + 1; 3869 APInt::tcSet(significandParts(), 0, partCount()); 3870 } 3871 3872 void 3873 APFloat::makeZero(bool Negative) { 3874 category = fcZero; 3875 sign = Negative; 3876 exponent = semantics->minExponent-1; 3877 APInt::tcSet(significandParts(), 0, partCount()); 3878 } 3879